Where Do Interorganizational Networks
Come From?

1

Ranjay Gulati
Northwestern University

Martin Gargiulo
INSEAD

Organizations enter alliances with each other to access critical re-
sources, but they rely on information from the network of prior alli-
ances to determine with whom to cooperate. These new alliances
modify the existing network, prompting an endogenous dynamic be-
tween organizational action and network structure that drives the
emergence of interorganizational networks. Testing these ideas on
alliances formed in three industries over nine years, this research
shows that the probability of a new alliance between specific organi-
zations increases with their interdependence and also with their
prior mutual alliances, common third parties, and joint centrality in
the alliance network. The differentiation of the emerging network
structure, however, mitigates the effect of interdependence and en-
hances the effect of joint centrality on new alliance formation.

INTRODUCTION

Sociologists have made considerable progress in explaining why organiza-
tions behave as they do in terms of their embeddedness in social networks
(Granovetter 1985, 1992; Swedberg 1994; Powell and Smith-Doerr 1994),
but they have seldom examined how those networks originated. With few
exceptions, largely limited to the research on interlocking directorates

1

We thank Steve Andrews, Wayne Baker, Joel Baum, Ron Burt, Bruce Carruthers,

Joe Galaskiewicz, Paul Hirsch, Linda Johanson, Mark Mizruchi, Joel Podolny, Haya-
greeva Rao, Andrej Rus, Dick Scott, Harrison White, and AJS reviewers for their
helpful comments on earlier drafts. Gulati thanks the Kellogg Graduate School of
Management, the Sloan Foundation Consortium for Competitiveness and Coopera-
tion, and the Harvard MacArthur Foundation for financial support. Gargiulo thanks
INSEAD and the Columbia University Graduate School of Business, where he was
a visitor in 1997–98, for financial support. Direct correspondence to Ranjay Gulati,
Kellogg Graduate School of Management, Northwestern University, Department of
Organization Behavior, 2001 Sheridan Road, Evanston, Illinois 60208-2001.



1999 by The University of Chicago. All rights reserved.

0002-9602/99/10405-0005$02.50

AJS Volume 104, Number 5 ( March 1999): 1439–93

1439

American Journal of Sociology

(e.g., Useem 1984; Palmer, Friedland, and Singh 1986; Mizruchi and
Stearns 1988) or to thick historical accounts of the development of particu-
lar interorganizational networks (e.g., Stern 1979), organizational sociolo-
gists have typically viewed network formation as driven by exogenous
factors, such as the distribution of technological resources or the social
structure of resource dependence (Pfeffer and Salancik 1978; Burt 1983).
In this view, organizations create ties to manage uncertain environments
and to satisfy their resource needs; consequently, they enter ties with other
organizations that have resources and capabilities that can help them cope
with these exogenous constraints.

The exogenous approach to tie formation provides a good explanation

of the factors that influence the propensity of organizations to enter ties,
but it overlooks the difficulty they may face in determining with whom
to enter such ties. This difficulty, which results from the challenges associ-
ated with obtaining information about the competencies, needs, and reli-
ability of potential partners (Van de Ven 1976; Stinchcombe 1990), is espe-
cially vivid in the case of interorganizational strategic alliances. Alliances
are a novel form of voluntary interorganizational cooperation that in-
volves significant exchange, sharing, or codevelopment and thus results in
some form of enduring commitment between the partners. While strategic
alliances can be a means to manage environmental uncertainty, there is
also considerable uncertainty associated with entering those cooperative
ties. Imperfect information about potential partners raises search costs
and the risk of exposure to opportunistic behavior (Gulati 1995a; Gulati
and Singh 1999). Thus, while exogenous factors may suffice to determine
whether an organization should enter alliances, they may not provide
enough cues to decide with whom to build those ties. Where do organiza-
tions find those cues? And how do the particular cues they use shape the
formation of interorganizational networks?

These are the two questions addressed in this article. We propose that

to reduce the search costs and to alleviate the risk of opportunism associ-
ated with strategic alliances, organizations tend to create stable, preferen-
tial relationships characterized by trust and rich exchange of information
with specific partners (Dore 1983; Powell 1990). Over time, these “embed-
ded” relationships (Granovetter 1985) accumulate into a network that be-
comes a growing repository of information on the availability, competen-
cies, and reliability of prospective partners (Kogut, Shan, and Walker
1992; Gulati 1995b; Powell, Koput, and Smith-Doerr 1996). The more the
emerging network internalizes information about potential partners, the
more organizations resort to that network for cues on their future alliance
decisions, which are thus more likely to be embedded in the emerging
network. These new embedded alliances, in turn, further increase the in-
formational value of the network, enhancing its effect on subsequent alli-

1440

Networks

ance formation. In this iterative process, new partnerships modify the pre-
vious alliance network, which then shapes the formation of future
cooperative ties. Thus, we model the emergence of alliance networks as
a dynamic process driven by exogenous interdependencies that prompt
organizations to seek cooperation and by endogenous network embed-
dedness mechanisms that help them determine with whom to build part-
nerships. Interorganizational networks are thus the evolutionary products
of embedded organizational action in which new alliances are increasingly
embedded in the very same network that has shaped the organizational
decisions to form those alliances.

We develop a model by specifying the mechanisms through which the

existing alliance network enables organizations to decide with whom to
build new alliances, and we discuss how the newly created ties can in-
crease the informational content of the same alliance network, enhancing
its potential to shape future partnerships. In theoretical terms, this is akin
to specifying the mechanisms through which social structures shape orga-
nizational action and the mechanisms through which this action subse-
quently affects social structures (Wippler and Lindenberg 1987; Gargiulo
1998). We test this model using longitudinal data on interorganizational
strategic alliances in a sample of American, European, and Japanese busi-
ness organizations in three different industries over a 20-year period. The
quantitative data collection and the empirical analysis for this study were
preceded by extensive interviews with managers involved in alliance deci-
sions at a variety of organizations. We conducted exploratory, open-ended
field interviews with 153 managers actively involved in alliance decisions
in 11 large multinational corporations. This fieldwork enabled us to
ground our claims about the role of the existing alliance network as a
source of information for organizational decision makers, as well as to
identify some of the mechanisms through which they tap that information.

INTERORGANIZATIONAL STRATEGIC ALLIANCES

Strategic alliances are a vivid example of voluntary cooperation in which
organizations combine resources to cope with the uncertainty created by
environmental forces beyond their direct control. These alliances are orga-
nized through a variety of contractual arrangements, ranging from equity
joint ventures to arm’s-length contracts (Harrigan 1986; Gulati 1995a;
Gulati and Singh 1999). Partly as a response to the growing uncertainty
that characterizes the international business arena, the number of interor-
ganizational alliances has grown at an unprecedented rate in the last 15
years, across a wide array of industries and both within and across geo-
graphical boundaries. Empirical evidence suggests that the number of in-
terorganizational alliances prior to 1980 was very small, but there has

1441

American Journal of Sociology

been a virtual explosion since that time (e.g., Hergert and Morris 1988).
The rapid growth of such ties provides a unique context in which to study
the emergence and the evolution of an interorganizational network from
the early stages of its development to the period in which alliances became
a more established form of cooperation among firms (Gulati 1998).

Despite their explosive growth, strategic alliances are associated with

a variety of risks and pitfalls that result in considerable uncertainty about
the decision to enter such ties. This is further compounded in the global
setting, with disparate firms from a wide range of national origins, in
which a good number of these alliances take place (Kogut 1988; Doz 1996).
This uncertainty stems from two main sources. First, organizations have
difficulty in obtaining information about the competencies and needs of
potential partners. This knowledge is essential to assess the adequacy of
a potential partnership if both organizations are to derive benefits from
the alliance. An organization that knows about the competencies and
needs of a potential partner is in a better position to assess whether the
alliance can simultaneously serve its own needs and its partner’s needs.
Yet organizational needs and capabilities are multifaceted and ambigu-
ous. Accurate information on needs and capabilities of other organizations
may be difficult to obtain before an alliance is initiated. In most cases, it
may require access to confidential information that would not be revealed
outside an established partnership. Such a paucity of information is even
more significant between firms from different geographic origins.

The second source of uncertainty that affects strategic alliances stems

from the paucity of information about the reliability of the potential part-
ners, whose behavior is a key factor in the success of an alliance (Gulati
1995a, 1995b). Such behavioral uncertainty is intrinsic to voluntary coop-
eration; indeed, it plays a central role in Coase’s ([1937] 1952) theory of
the firm and in the transaction-cost perspective (Williamson 1985). Orga-
nizations entering alliances face considerable moral hazard concerns be-
cause of the unpredictability of the behavior of partners and the likely
costs to an organization from opportunistic behavior by a partner, if it
occurs (Kogut 1988; Doz, Hamel, and Prahalad 1989). A partner organiza-
tion may either free ride by limiting its contributions to an alliance or may
simply behave opportunistically, taking advantage of the close relation-
ship to use resources or information in ways that may damage the part-
ner’s interests. In addition, rapid and unpredictable changes in the en-
vironment may lead to changes in an organization’s needs and its
orientation toward ongoing partnerships (MacIntyre 1981).

The paucity of reliable information about the capabilities, the needs,

and the behavior of potential partners creates a significant informational
hurdle for organizations that consider entering strategic alliances. Yet the

1442

Networks

explosive growth of strategic alliances suggests that organizations are able
to overcome such hurdles and enter alliances. How do they do it? And
what consequences does their behavior have for the social context in
which new strategic alliances take place?

THE FORMATION OF STRATEGIC ALLIANCES

Interdependence

Interdependence is the most common explanation for the formation of
interorganizational cooperative ties such as strategic alliances. A long
stream of research suggests that organizations enter ties with other organi-
zations in response to the challenges posed by the interdependencies that
shape their common environment (e.g., Aiken and Hage 1968; Pfeffer and
Salancik 1978; Aldrich 1979; Galaskiewicz 1982; Burt 1983). Broadly de-
fined, environmental dependence encompasses two sets of considerations:
resource procurement and uncertainty reduction (Galaskiewicz 1985). Or-
ganizations build cooperative ties to access capabilities and resources that
are essential to pursue their goals but that are at least in part under the
control of other organizations in their environment. Interorganizational
cooperation is thus a means by which organizations manage their depen-
dence on other organizations in their environment and attempt to mitigate
the uncertainty generated by that dependence. Oliver (1990) reviewed the
literature on such exogenous drivers of interorganizational relations and
presented six broad categories of environmental contingencies that stimu-
late such ties. One of these types of contingencies—necessity—prompts
ties mandated by legal or regulatory requirements, but the other catego-
ries—asymmetry, reciprocity, efficiency, stability, and legitimacy—lead
to cooperative ties that organizations voluntarily initiate to address spe-
cific needs resulting from their external interdependence.

Strategic alliances are an important form of voluntary cooperative in-

terorganizational ties. Organizations build alliances for a variety of rea-
sons, including the need to share the costs and risks of technology develop-
ment or large-scale projects, to develop existing markets or penetrate new
ones, and to pursue resource specialization strategies (Mariti and Smiley
1983). Such objectives make organizations interdependent with other or-
ganizations that may have the capabilities and the resources to assist them
in meeting their specific needs. Other things being equal, the higher the
interdependence between two organizations, the higher their incentive to
combine their resources and capabilities through an alliance. Building on
the insights of this research tradition, we expect tie formation between or-
ganizations to be a function of the level of interdependence between them.
Thus:

1443

American Journal of Sociology

Hypothesis 1.—The probability of a new alliance between two organi-

zations increases with the level of interdependence between those organi-
zations.

Interdependence may be a necessary condition for organizations to en-

ter alliances. In most cases, however, interdependence may not be suffi-
cient to account for the formation of an alliance between two specific
firms. Indeed, not all opportunities for cooperation between interdepen-
dent organizations actually materialize in alliances. An organization con-
fronted with the need to build an alliance to cope with an uncertain envi-
ronment faces another type of uncertainty resulting from the identification
of an appropriate alliance partner. Such uncertainty stems from the pau-
city of information about the true capabilities, the needs, and the behavior
of potential alliance partners. While interdependence may help an organi-
zation to orient the search for an adequate alliance partner, it cannot offer
sufficient cues to determine with whom it should build such an alliance.
This has not posed major difficulty for studies conducted at aggregate
levels of analysis, which typically predict the formation of alliances across
industries as a function of the intensity of the transactions between those
industries (Pfeffer and Nowak 1976a, 1976b; Berg and Friedman 1980;
Duncan 1982; Burt 1983). Yet, this approach masks the considerable het-
erogeneity of available information on prospective partners across organi-
zations, which may influence the formation of ties between specific organi-
zations without necessarily affecting aggregate industry trends. Although
resource-dependence perspectives recognize an “enactment” process that
mediates between environmental demands and organizational action
(Pfeffer and Salancik 1978, p. 71), most of this research implicitly assumes
that decision makers have adequately identified the sources of environ-
mental uncertainty as well as the partners that would help their organiza-
tions to reduce that uncertainty. While this assumption is tenable at aggre-
gate levels of analysis, it is difficult to sustain when examining alliances
between specific pairs of organizations.

Interorganizational Embeddedness

If interdependence alone cannot offer sufficient cues for organizations to
cooperate with one another, how do they decide with whom to build stra-
tegic alliances? Building on a growing body of research (see Powell and
Smith-Doerr [1994] and Gulati [1998] for a review) and on our own field-
work, we argue that organizations address the potential hazards associ-
ated with building alliances by relying on information provided by ex-
isting interorganizational networks. We propose that organizational
decision makers that play a crucial role in the formation of new strategic
alliances rely on the network of past partnerships to guide their future

1444

Networks

alliance decisions. The creation of new ties, in turn, contributes to the
subsequent development of that same network, enhancing its capacity to
shape subsequent alliance decisions.

Although rooted in classical sociological theory, the idea that economic

action is embedded in social networks was revitalized by Granovetter
(1985) in his manifesto for a new economic sociology. According to Grano-
vetter (1985, p. 490), the microfoundations of embedded economic action
rest on “the widespread preference for transacting with individuals of
known reputation,” for resorting to “trusted informants” who have dealt
with a potential partner and found this partner trustworthy, or, even bet-
ter, for relying on “information from one’s own past dealings with that
person.” A similarly rich exchange of information occurs across organiza-
tional boundaries (Dore 1983; Eccles 1981; Powell 1990; Romo and
Schwartz 1995). Personal relationships among key individuals have
played a crucial role in producing trust between organizations in Japanese
industrial groups (Lincoln, Gerlach, and Ahmadjian 1996) and in contrac-
tual relationships (Macaulay 1963; Bradach and Eccles 1989). Closer to
our concerns, personal ties are important for the formation and success
of strategic alliances (Ring and Van de Ven 1992; Doz 1996). Beneath
the formalities of contractual agreements, multiple informal interpersonal
relationships emerge across organizational boundaries, which facilitates
the active exchange of information and the production of trust that foster
interorganizational cooperation (Gulati 1995a; Walker, Kogut, and Shan
1997; Zaheer, McEvily, and Perrone 1998).

Most organizations are embedded in a variety of interorganizational

networks, such as board interlocks, trade associations, and research and
development ventures. Scholars have suggested that participation in such
social networks can be influential in providing actors with access to timely
information and referrals to other actors in the network (Burt 1992). At
the interorganizational level, the network of prior alliances has been iden-
tified as one such network that is an important source of information and
referrals for organizations (Kogut, Shan, and Walker 1992; Gulati 1995b).
This insight was strongly confirmed by managers in our own fieldwork
who highlighted the importance of the network of prior alliances as a
source of trustworthy information about the availability, capabilities, and
reliability of potential partners. In the words of one of the managers inter-
viewed, “Our network of [prior alliance] partners is an active source of
information for us about new deals [alliances]. We are in constant dialogue
with many of our partners, and this allows us to find many new opportuni-
ties with them and also with other organizations out there.”

The information that flows through the alliance network is not only

trustworthy but is also timely. This, as another manager put it, is critical
for entering strategic alliances: “In our business, timing is everything. And

1445

American Journal of Sociology

so, even for alliances to happen, the confluence of circumstances have to
be at the right time. We and our prospective partner must know about
each other’s needs and identify an opportunity for an alliance together in
a timely manner. . . . Our partners from past alliances are one of our most
important sources of timely information about alliance opportunities out
there, both with them and with other firms with whom they are ac-
quainted.”

Existing network research and insights from our own fieldwork suggest

that timely, relevant information on the competencies, needs, and reliabil-
ity of potential partners originates from organizations’ previous direct alli-
ances, from their indirect alliance ties through third parties, or from the
reputation that results from the potential partner’s position in the network
of preexisting alliances. Each of these sources of information is related to
specific network mechanisms that shape the creation of new embedded
interorganizational ties. We refer to these mechanisms as relational, struc-
tural, and positional embeddedness, respectively.

Relational embeddedness highlights the effects of cohesive ties between

social actors on subsequent cooperation between those actors. Cohesive
ties play a prominent explanatory role in classical sociological analysis of
social solidarity and cooperation (e.g., Durkheim [1893] 1933, [1897] 1951).
Prior cohesive ties between two organizations provide channels through
which each partner can learn about the competencies and the reliability
of the other. Cohesiveness amplifies trust and diminishes the uncertainty
associated with future partnerships (Podolny 1994; Burt and Knez 1995;
Gulati 1995a). Cohesive ties may also prompt organizations to become
aware of new opportunities for cooperation that would be difficult to iden-
tify outside of a close relationship. This facet of cohesive relationships
was emphasized by one of the strategic-alliance managers we interviewed:
“They [our partners] are familiar with many of our projects from their
very inception, and if there is potential for an alliance we discuss it. Like-
wise, we learn about many of their product goals very early on, and we
actively explore alliance opportunities with them.” Thus, a history of co-
operation can become a unique source of information about the partner’s
capabilities and reliability and increases the probability of the two organi-
zations forming new alliances with each other. As a consequence:

Hypothesis 2.—The probability of a new alliance between two organi-

zations increases with the number of prior direct alliances between those
organizations.

Structural embeddedness captures the impact of the structure of rela-

tions around actors on their tendency to cooperate with one another (Gra-
novetter 1992). The frame of reference shifts from the dyad to the triad,
while the focus of analysis shifts from direct communication between

1446

Networks

actors to indirect channels for information and reputation effects.

2

Organi-

zations tied to a common partner can utilize reliable information about
each other from that partner (Baker 1990; Gulati 1995b). When two orga-
nizations share common ties, it can also indicate that both are regarded
as suitable and trustworthy by the same organizations. Also, sharing com-
mon ties with a potential partner may signal that the partner can cooper-
ate with the same kind of organizations with which the focal organization
has been cooperating. Finally, common third-party ties can create a repu-
tational lock-in whereby good behavior is ensured through a concern for
local reputation. Any bad behavior by either partner may be reported to
common partners, which serves as an effective deterrent for both (Raub
and Weessie 1990; Burt and Knez 1995).

Referrals, and their associated reputation effects, were explicitly men-

tioned in several of our field interviews as an important mechanism
through which their organizations learned about reliable partners. In the
words of one of the managers, “In some cases we realize that perhaps our
skills do not really match for a project, and our partner may refer us to
another organization about whom we were unaware. . . . An important
aspect of this referral business is of course about vouching for the reliabil-
ity of that organization. Thus, if one of our long-standing partners suggests
one of their own partners as a good fit for our needs, we usually consider
it very seriously.” Thus:

2

Although there is no explicit mention by Granovetter (1992), the notion of structural

embeddedness is related to network models of structural equivalence, according to
which two actors equally tied to the same third parties are “structurally equivalent”
(Lorrain and White 1971; White, Boorman, and Breiger 1976; Burt 1976). There has
been considerable debate about whether structural equivalence operates through the
indirect effect of cohesive ties to common third parties (Alba and Kadushin 1976; Alba
and Moore 1983; Friedkin 1984) or through “symbolic role playing” and competition
between equivalent actors (Burt 1982, 1987). The essence of the debate is whether
the mechanisms behind the impact of indirect ties on behavior are substantially differ-
ent from those behind the effect of direct ties. Yet, as Borgatti and Everett (1994, p.
28–29) have suggested, the notion of proximity is an inseparable part of structural
equivalence. In a similar vein, Mizruchi (1993, p. 280) suggested that deciding whether
the effects of structural equivalence on behavior operate through the similar socializ-
ing pressures of common third parties or through symbolic role playing is practically
impossible without knowing the motives that underlie the actors’ behavior. It is worth
noting, however, that a mechanism that stresses competition between structurally
equivalent actors (e.g., Burt 1987) would predict a smaller probability of cooperation
between actors tied to the same third parties, whereas a mechanism that stresses the
increased trust and information between those with common third-party ties predicts
a greater probability of cooperation between the actors (e.g., Coleman 1990). Our focus
on the increased trust and information effects of third-party ties is congruent with
our prediction that shared partners increase the probability of cooperation between
organizations.

1447

American Journal of Sociology

Hypothesis 3.—The probability of a new alliance between two organi-

zations increases with the number of prior indirect alliances between those
organizations.

Positional embeddedness captures the impact of the positions organiza-

tions occupy in the overall structure of the alliance network on their deci-
sions about new cooperative ties. Positional embeddedness is rooted in
network models of equivalence and centrality that capture the “roles”
actors occupy in a system, irrespective of the specific alters involved in
playing those roles (Winship and Mandel 1983; Faust 1988; Borgatti and
Everett 1994). As a mechanism that influences alliance formation, posi-
tional embeddedness goes beyond proximate direct and indirect ties and
highlights the informational benefits that ensue for organizations from
particular positions in the network. The position an organization occupies
in the emerging network can influence its ability to access fine-grained
information about potential partners as well as its visibility and attrac-
tiveness for other organizations throughout the network, even if it is not
directly or indirectly tied to them. The information advantages resulting
from network centrality have been a recurrent theme in network analysis
(see Freeman [1979] for a review). Social cognition studies also suggest
that central actors have a more accurate representation of the existing
network (Krackhardt 1990). Central organizations have a larger “intelli-
gence web” through which they can learn about collaborative opportuni-
ties, hence lowering their level of uncertainty about partnerships (Gulati
1999; Powell et al. 1996). Therefore, the more central an organization’s
network position, the more likely it is to have better information about a
larger pool of potential partners in the network.

The information advantages from centrality in networks are comple-

mented by the higher visibility of central organizations, which enhances
their attractiveness to potential partners. Because network centrality is a
direct function of organizations’ involvement in strategic alliances, it can
also be a signal of their willingness, experience, and ability to enter such
partnerships. The signaling property of network positions is particularly
important in uncertain environments, because it introduces systemic repu-
tational differences among organizations that extend beyond their imme-
diate circle of direct and indirect ties (Podolny 1993; Han 1994; Podolny
and Stuart 1995).

3

The information benefits that result from occupying a

3

Network position often has been associated with the more traditional sociological

concepts of “role” and “status” (Lorrain and White 1971; Burt 1982; Faust 1988). The
notion of “role” typically evokes a relatively defined set of expected behaviors toward
types of other actors, whereas “status” refers to a series of observable characteristics
associated with a particular role (Linton 1936; Merton 1957; Nadel 1957). Network
theory suggests that because an actor’s (organization’s) role and status are ultimately

1448

Networks

prominent network position were recognized by a manager we inter-
viewed who reflected on the attractiveness of his firm as alliance partner:
“Through our vastly successful technology partnerships program, we have
built ourselves a reputation in the industry for being an effective and reli-
able alliance partner. Today, we are pursued by other firms to enter alli-
ances much more frequently than we pursue potential partners.” If central
firms have greater access to information and higher visibility than other
organizations, then, other things being equal, interorganizational ties
should be more common between organizations that occupy central posi-
tions in the emerging interorganizational network. Thus:

Hypothesis 4.—The probability of a new alliance between two organi-

zations increases with the combined alliance-network centrality of those
organizations.

Organizations may seek to enhance their own visibility and attrac-

tiveness as potential partners by forming new ties with central players
in the alliance network. Since the network position of an organization’s
partners enhances its own access to information and attractiveness to fu-
ture partners, it will have a tendency to seek central partners. Central
organizations, however, may not have an incentive to accept peripheral
players, since they may add little to (or, worse, may damage) their own
attractiveness. Furthermore, if network position is a signal of unobserved
attributes that determine an organization’s attractiveness as a potential
alliance partner, peripheral organizations may be perceived by others to
have little to offer substantively. This does not preclude the possibility
that peripheral organizations may at times enter alliances with central
firms. Special circumstances such as those resulting from the need to mas-
ter a new technology may prompt a central organization to cooperate with
a peripheral one that controls such a technology, but we expect that the
probability of cooperation will increase with the similarity in alliance-
network centrality between the potential partners. Therefore:

Hypothesis 5.—The probability of a new alliance between two organi-

zations increases with the similarity in alliance-network centrality be-
tween those organizations.

The prediction of this hypothesis corresponds to the tendency toward

“structural homophily” that exist under conditions of uncertainty (Podolny
1994; Popielarz and McPherson 1995).

THE ENDOGENOUS DYNAMIC OF ALLIANCE NETWORKS

In building new alliances, organizations also contribute to the formation
of the network structure that shapes future partnerships. Observed over

based on its affiliations and patterns of interaction, they can, and should be, gauged
from the position the actor occupies in the networks defining the social system.

1449

American Journal of Sociology

time, this dynamic between embedded organizational action and the net-
work structure that results from that action propels the progressive struc-
tural differentiation of the interorganizational network. We define struc-
tural differentiation as an emergent systemic property that captures the
extent to which actors (organizations) come to occupy an identifiable set
of network positions, each of them characterized by a distinctive relational
profile. Because the position an organization occupies in an alliance net-
work is a signal of its willingness, experience, and ability to enter partner-
ships, the higher the structural differentiation of an emerging network,
the easier it is for organizations to distinguish among other organizations
in terms of their relational profiles and the more the network structure
acts as a repository of valuable information on potential alliance partners.

This discussion suggests a linear relationship between the level of struc-

tural differentiation of the emerging alliance network and the extent of
information contained in that network. There is, however, an important
caveat. While a network in which all or most organizations have a similar
relational profile would offer little guidance to a decision maker, the oppo-
site extreme, a network in which each organization has a truly unique
relational profile, may be equally uninformative. This is particularly sig-
nificant for the information that originates in the position an organization
occupies in the structure of the alliance network. If every organization
were to occupy a unique structural position, it would be impossible to infer
the behavior of any particular organization from the expected behavior of
other organizations that occupy that position in the system. The underly-
ing social structure would offer little guidance to organizations seeking
an alliance partner, since every potential partner would be unique from
a network standpoint. As a result, the relationship between the structural
differentiation of a network and the information available to the actors
in that network may level off and eventually even become negative as
the accumulation of new ties further increases the level of structural differ-
entiation in the network beyond a critical level. Studies of mature social
structures suggest that the structural differentiation of most real systems
may not display a continuous increase over time. Instead, mature struc-
tures typically display a set of stable, self-reproducing positions occupied
by actors with similar network profiles (White 1981; Burt 1988). In such
structures, the level of structural differentiation remains practically con-
stant over time. Barring exogenous shocks, the structural differentiation
of alliance networks may similarly taper off as the social structure of the
interorganizational network reaches a mature state.

The effects of structural differentiation are conceptually distinct from

the legitimating effects typically associated with growing network density
(Hannan and Freeman 1989; Scott 1995). Although structural differentia-
tion is likely to grow with the number of ties in the network, it is distinct

1450

Networks

insofar as it depends on the specific distribution of those new ties, not
merely on their number. The density of ties in a network may provide
organizations with information about the pervasiveness of a new form of
cooperation, thus helping them to address concerns on the legitimacy of
this course of action, but it offers no guidance as to which specific organi-
zations could be worthy partners. Thus, while network density affects the
availability of information in a system (Blau 1977), it does not shed light
on potential differences in effective access to that information, nor on how
the pattern of ties may themselves provide information.

We expect that the structural differentiation of the emerging alliance

network will influence new alliance formation both directly and through
its interaction with some of the mechanisms that drive alliance formation.
At the system level, the additional information introduced by the progres-
sive structural differentiation of an emerging network lowers the level
of systemic uncertainty faced by organizations, which directly affects the
propensity of organizations to enter new ties. Thus:

Hypothesis 6.—The probability of a new alliance between any two

organizations increases with the level of structural differentiation in the
interorganizational network.

While structural differentiation focuses on system-level information, ex-

ogenous resource concerns and network embeddedness focus on the more
proximate level of organizations. Given the shared focus of these factors
on information availability, we expect that an increase in the extent of
structural differentiation is likely to moderate the relative influence of
interdependence and embeddedness factors on the creation of new ties.
In early periods, when a network is relatively undifferentiated and thus
likely to contain limited information about potential partners, organiza-
tions may still be prompted to cooperate by exogenous pressures that in-
fluence their interests. As a consequence, exogenous factors are likely to
be the primary driver of tie formation in the early stages of a network, but
the growing differentiation of a network enables it to channel increasing
amounts of information about potential partners. As structural differenti-
ation increases, exogenous factors are likely to have a diminishing influ-
ence on the formation of new ties. Thus, we expect the structural differen-
tiation of the network to have a negative moderating effect on the
influence of exogenous factors on tie formation:

Hypothesis 7.—The effect of interdependence on the formation of new

alliances between organizations decreases with the level of structural dif-
ferentiation of the interorganizational network.

We also expect the structural differentiation of the network to moderate

the influence of embeddedness on tie formation, although not all embed-
dedness mechanisms are likely to be moderated by the growing differenti-
ation of the network. The information organizations can obtain through

1451

American Journal of Sociology

previous direct dealings with other organizations (relational embed-
dedness) or from common third-party alliances (structural embeddedness)
is readily available to a decision maker, and thus it is not necessarily de-
pendent on the larger network in which these dyadic or triadic relations
exist. Access to such information depends on the ability of proximate ties
to act as conduits of fine-grained information about the competencies and
cooperative behavior of other organizations, a property that is not contin-
gent on the stage of development of the entire network. Therefore, the
impact of relational- and structural-embeddedness mechanisms is not nec-
essarily contingent on the level of structural differentiation in the overall
network.

4

While the information that results from prior ties to a prospective part-

ner or from common third parties is immediately available to organiza-
tions, this is not the case with the information contained in the position
their potential partners occupy in the emerging alliance network. The ef-
fectiveness of an organization’s network position as a signal of unobserv-
able qualities of this organization depends on the development of the
overall network in which the varying involvement of organizations in
partnerships becomes apparent. The relative scarcity of ties at early stages
of the network makes these differences far from apparent. The increase
in structural differentiation corresponds to an increase in differences in
alliance involvement across organizations, which alters their relative visi-
bility in the overall network. Thus, the informational value of the position
of organizations in a social network is likely to be contingent on the level
of structural differentiation of that network. We therefore expect the effect
of organizations’ positional embeddedness on tie formation to increase
with the level of structural differentiation of the network:

Hypothesis 8.—The effect of positional embeddedness on the forma-

tion of new alliances between organizations increases with the level of
structural differentiation of the interorganizational network.

Figure 1 summarizes our dynamic model of network formation and

highlights the empirically testable predictions of the model. The solid

4

Our reluctance to suggest that the effects of relational or structural embeddedness

are contingent on structural differentiation does not rule out alternative mechanisms
through which the growth of the network may alter the effect of these factors and
perhaps lead to an empirically observable relationship. The sheer growth in network
density could enhance the legitimacy of partnerships, thus making organizations more
eager to build ties. Insofar as we expect organizations to prefer embedded ties, the
likelihood of entering new ties with previous partners or with common third parties
may increase with the growing density of the network. Since density is a likely corre-
late of structural differentiation, one may still observe a growing impact of relational
embeddedness as differentiation increases, but this effect is likely to be spurious from
the standpoint of our model.

1452

Networks

Fig. 1.—The endogenous dynamic of interorganizational networks

arrows represent the direct effects of the key variables on network forma-
tion (strategic interdependence, relational, structural, and positional em-
beddedness, and structural differentiation). The dotted arrows from struc-
tural differentiation to the arrows for the direct effects of interdependence
and positional embeddedness capture the moderation effect of structural
differentiation on the impact of those mechanisms on tie formation. The
plus and minus signs indicate a strengthening or weakening of influence
in the direction of the arrows. Our expectation is that the greater the struc-
tural differentiation of the emerging network, the stronger the effects of
positional embeddedness and the weaker the effect of strategic interdepen-
dence. The dashed arrow from network formation to structural differenti-
ation indicates the dynamic connection between action and structure.

METHODS

Sample

We tested our model using longitudinal data on strategic alliances in a
sample of American, European, and Japanese organizations in three in-

1453

American Journal of Sociology

dustries over a 20-year period. We collected data on a sample of 166
organizations in new materials, industrial automation, and automotive
products. We selected a panel of 50–60 of the largest publicly traded
organizations within each sector, estimating an organization’s size from
its sales in that sector as reported in various industry sources. We also
checked with multiple industry experts to ensure that our panels included
all prominent competitors in the sectors. This design led to the inclusion
of 62 organizations in new materials, 52 in automotive products, and 52
in industrial automation. Of these organizations, 54 were American, 66
were Japanese, and 46 were European.

For each organization, we collected financial data for each year between

1980 and 1989 from Worldscope, which provides detailed information
about prominent organizations in a wide range of sectors. For organiza-
tions not reported in Worldscope, data were obtained from COMPUSTAT
for U.S. organizations, Nikkei for Japanese organizations, and Disclosure
for European organizations. For a number of Japanese organizations, data
were also obtained from Daewoo Investor’s Research Guide.

5

We also

collected information for each organization from numerous industry-spe-
cific trade journals about the subsegment of its industry within which it
had expertise. To make sure that these classifications were correctly re-
corded, we cross-checked these with multiple experts from each of the
industries.

Information on the alliances formed in the three panels of organizations

was derived from a much larger and more comprehensive data set that
includes information on over 2,400 alliances formed by American, Euro-
pean, and Japanese organizations in the three focal sectors for 1970–89.
More than half the data came from the Cooperative Agreements and
Technology Indicators (CATI) database collected by researchers at the
University of Maastricht. We collected additional alliance data from nu-
merous other sources, including industry reports and industry-specific ar-

5

For a few organizations, financial data were available for only some years. The gaps

typically resulted from the fact that Worldscope reports organization data in five-
year continuous segments and omits some organizations from some volumes. One
alternative for dealing with this problem would have been to use the “available-case
method,” including only cases with the variables of interest in the analysis. Although
such an approach is straightforward, it poses a number of problems, including vari-
ability in the sample base as the variables included in models change. Furthermore,
it makes little sense to exclude entire cases simply because a single variable is missing.
We thus chose to estimate the missing data using a time-trend-based imputation (Little
and Rubin 1987). This procedure took into account the fact that the financial outcome
for an organization is the result of its own past actions as well as broad trends within
its industry. We retained a dummy variable indicating imputation and later compared
the results obtained with and without imputed values.

1454

Networks

ticles reporting alliances. For the automotive industry, these included Au-
tomotive News, Ward’s Automotive Reports, U.S. Auto Industry Report,
Motor Industry of Japan, and the Japanese Auto Manufacturers Forum;
for the industrial automation sector, Managing Automation (1988–89); for
the new materials sector, reports from the Office of Technology Assess-
ment and the Organization for Economic Cooperation and Development
were used; and for all sectors, we used Predicast’s Funk and Scott Index
of Corporate Change. In all instances, we recorded only alliances that had
actually been formed and excluded reports of probable alliances. To our
knowledge, these are the most comprehensive data on alliances within
each focal sector in both depth and duration of coverage.

The Structure of Interorganizational Alliance Networks

We analyzed the networks in each industry in our sample to explore the
structure of the alliance networks and visually examined the emergence
of structurally differentiated positions to assess the structural patterns that
would clarify and illustrate the differentiation process depicted in the the-
oretical model. We examined the network structure resulting from the
cumulative alliance activity of the organizations within each industry by
conducting separate analyses of each industry network in the penultimate
period of the study (1988). Each of the three networks included all interor-
ganizational alliances announced in that industry in the previous five
years (1983–88). The strength of the ties between two organizations in the
network corresponded to the strongest alliance between these organiza-
tions in the period, where strength is measured on a seven-point Guttman
scale (Contractor and Lorange 1988; Nohria and Garcia-Pont 1991). We
used the concept of “role equivalence” to identify classes of organizations
or “positions” in the network and their relationship. Role equivalence cap-
tures similarities in the organizations’ pattern of involvement in alliances,
even when this involvement may be with different partners. Each role-
equivalent position refers to sets of actors involved in similar types of
relations but not necessarily with the same “alters.” While structural-
equivalence models focus on relations with specific actors (Lorrain and
White 1971), role-equivalence models focus on the pattern of relationships
among actors and are more adequate to capture status/role sets in a net-
work (Winship and Mandel 1983; Faust 1988; Borgatti and Everett 1994).
Two actors are structurally equivalent if they have similar relationships
with similar alters, while two actors are role equivalent when they are
involved in similar types of relationships with others actors. For instance,
two managers leading separate divisions are not structurally equivalent
because they have different subordinates; however, they are role equiva-

1455

American Journal of Sociology

lent, since they have a similar type of relationship with these subordi-
nates.

6

To identify role-equivalent positions in the interorganizational net-

works, we used an approach developed by Hummel and Sodeur (1987)
and by Burt (1990). Building on the triad-census idea introduced by Hol-
land and Leinhardt (1970), this technique identifies role equivalence in
terms of similarity in the actor’s triad patterns. The larger the extent to
which two actors are involved in similar triads, the more they are role
equivalent (see Van Rossem [1996] for an example of this application).

7

Role equivalence is measured as the euclidean distance between the vec-
tors that capture the triad pattern of each actor. Two organizations that
had identical triad patterns would be separated by zero euclidean dis-
tances and would be perfectly role equivalent, regardless of the specific
organizations with which they built their alliances.

We computed role-equivalence measures using the general-purpose net-

work analysis software Structure 4.2 (Burt 1991).

8

Role-equivalent blocks

were identified by cluster analysis of the matrix of euclidean distances
between the organizations’ triad patterns. Finally, we also performed met-
ric multidimensional-scaling analyses of the proximity matrices of the in-
dustry networks, in which proximity was defined by the strength of the
alliance between organizations.

Figure 2 presents density tables based on role-equivalent partitions of

the three industry networks, along with spatial maps of each industry

6

Several network analytical concepts—and algorithms—have been proposed to cap-

ture this abstract form of equivalence, including automorphic, regular, positional, and
role equivalence. For a comprehensive reviews of these concepts, see Mizruchi (1993),
Wasserman and Faust (1994), and Borgatti and Everett (1994).

7

In a symmetric network, a focal organization, or “ego,” can be involved in six differ-

ent triads with two other organizations, or “alters.” These triads are T1, in which all
three parties are disconnected—also known as the “null triad”; T2, in which ego is
connected to only one of two disconnected alters; T3, in which ego is connected to
two disconnected alters; T4, in which ego faces two connected alters but has no con-
nection to either; T5, in which ego is connected to one of two connected alters; and
T6, in which all three actors are connected. T1 and T4 define an isolated role, while
T3 is typical of central roles. The isolated triad (T1), however, has a disproportionately
high frequency in sparse networks such as the ones analyzed here. To eliminate poten-
tial biases that stem from this dominance, we excluded the null triad from the census
(Van Rossem 1996). The isolated role is thus purely defined by T2.

8

An alternative approach would have been to use a regular equivalence algorithm,

such as the one included in UCINET IV. The algorithm, however, has posed computa-
tional and interpretative difficulties when applied to symmetric networks (Doreian
1987, 1988; Borgatti 1988). The triad-census approach is computationally simpler and
has an intuitive appeal; it is similar to the original Winship and Mandel’s (1983) model
if role equivalence is defined by direct and two-step ties only (Burt 1990).

1456

Fig. 2.—Metric MDS maps and density tables of the interorganizational net-

work structures, 1989. Proximities in MDS maps are defined by the strength of
the alliances; positions are occupied by role-equivalent organizations. Density is
the mean strength of alliances between occupants of the respective position(s),
measured on a seven-point Guttman scale. Mean density is the value for the whole
industry network and equals .289 for automotive, .115 for industrial automation,
and .106 for new materials. Centr. equals the mean eigenvector centrality within
the position. Rel. equals the percentage of variance in euclidean distances ex-
plained by the first principal component of the position submatrices.

American Journal of Sociology

based on the first two dimensions of a metric multidimensional-scaling
analysis of the observed alliance networks. For all three industries, the
analysis clearly revealed four distinctive positions of role-equivalent orga-
nizations, one of which is occupied by “isolates”—organizations that did
not enter any alliance in the five years prior to 1989. Figures in the main
diagonal cells represent the average strength of an alliance between any
two organizations occupying the same structural position; figures in off-
diagonal cells represent the average strength of alliances between organi-
zations in the respective two positions. Since we measured alliance
strength on a seven-point Guttman scale, figures can vary from zero—
when there is no alliance between any two organizations occupying the
respective positions—to seven—if all organizations in the position(s) were
tied to one another through the strongest alliances. Because alliances are
symmetric ties, density tables are symmetric along the main diagonal. The
tables also report average eigenvector centrality scores for the organiza-
tions in each position.

9

We tested the homogeneity of the resulting posi-

tions with a principal component analysis of the submatrices of euclidean
distances between position occupants. In all but one of the cases, the first
principal component explained 90% or more of the observed variance in
distances within the position, which suggested that the positions were oc-
cupied by organizations that were strongly role equivalent.

An inspection of the three density tables and spatial maps revealed a

similar structural pattern across the three industries. First, the same num-
ber of positions adequately described all three interorganizational struc-
tures. The homogeneity of these positions was also consistent across indus-
tries. This suggested that all three industries experienced a similar pattern
of structural differentiation during our observation period and limits po-
tential concerns associated with industry differences.

Second, the figures in the cells revealed a uniform core-periphery pat-

tern. In all three industries, organizations in the central position or “core”
(position 1, represented by the black dots in the spatial maps) built strong
alliances with one another and somewhat weaker, but still significant,
alliances with organizations in the “semiperiphery” (position 2, gray dots)
and the “periphery” (position 3, empty dots). In addition, all structures
contain a considerable number of “isolates” (position 4, empty squares).
The average alliance between the core and semiperiphery was stronger
than that between the core and the periphery. The mean strength of intra-
position alliances for organizations in the semiperiphery and periphery—
displayed in the respective cell of the main diagonal—was always smaller

9

Centrality scores are the average eigenvector centrality (Bonacich 1987), normalized

by industry to vary between 0 and 1. Central organizations are involved in alliances
with partners who are in turn involved in many alliances.

1458

Networks

than the mean strength of alliances between those organizations and the
ones in the core—displayed in the first row (or column) of the tables. A
similar pattern existed for the alliances between the semiperiphery and
the periphery: except for the automotive industry, these alliances were
stronger than the alliances between organizations in the periphery. Thus,
periphery and semiperiphery organizations were more likely to build ties
with organizations in the center than with organizations occupying their
same positions. Core organizations, in turn, were more likely to build alli-
ances with other core organizations than with either semiperiphery or pe-
riphery organizations. This pattern is typical of core-periphery structures
(Van Rossem 1996).

Third, there was no evidence of isolated factions within the alliance

networks. An inspection of the density tables and the accompanying spa-
tial maps showed no indication of dense clusters of cohesive organizations
with little interaction outside the cluster. Further, structural analysis using
various clique detection routines also failed to identify isolated cohesive
subgroups. Although all three networks displayed a relatively cohesive
core of varying size, its occupants were also heavily involved in building
alliances with organizations in the semiperiphery and—to a lesser ex-
tent—with those in the periphery.

The emergence of a core-periphery structure, such as the one revealed

by our positional analysis, is consistent with the mechanisms in our endog-
enous embeddedness model. Organizations may originally differ in their
propensity to build ties because of variability in exogenous pressures or
organization-specific characteristics, yet the logic of relational and struc-
tural embeddedness amplify even a small initial variance in alliance activ-
ity, eventually creating appreciable differences among organizations. Re-
lational embeddedness suggests that the more active organizations should
have better information on potential partners, since they have access to
a larger number of previously trusted partners. Structural embeddedness
expands this pool to the partners’ partners, which are likely to increase
exponentially with the number of direct ties. Since having information on
a larger pool of comparatively trustworthy potential partners increases
the probability of entering new ties, the more active organizations have
increasing comparative informational advantages over less active ones,
which increases their likelihood of entering new partnerships (Gulati
1999). The differential involvement in partnerships eventually affects the
visibility of the most active organizations beyond the circle defined by
their direct and indirect ties, prompting further observable differences
among organizations in the emerging network. These differences create
conditions that can boost the influence of positional embeddedness on new
alliance formation, because they make it easier for organizations to recog-
nize central players in the emerging alliance network.

1459

American Journal of Sociology

Variables

Alliance formation.—Our dependent variable is the formation of a new

alliance between two organizations in a given observation year. Since the
unit of analysis was the dyad, for each panel we listed all possible dyads
within each sector, discarding reverse-ordered dyads to avoid double
counting.

10

These data were then used to construct an event history for

each dyad, with a record for each dyad for each year studied (1981–89).
For each dyad-year record, we coded a dichotomous dependent variable
that indicated whether the pair of organizations entered an alliance in the
given year. The resulting data structure is best characterized as a cross-
sectional time-series panel in which the units are unique dyads. Each rec-
ord included the state of the dependent variable, indicating the formation
of an alliance in that period, along with time-varying and time-constant
covariates characterizing the dyad.

Such a broad definition of the risk set, which included all possible dyads

for the sample of firms in each industry, was considered essential to uncov-
ering unbiased results. Including many dyads that never enter an alliance
can, of course, lead to its own set of biases, but we had no observable
criteria to determine a priori which dyads were likely to enter alliances
and which were not. To address this issue and test the robustness of the
results, we conducted the analysis with two additional risk sets that were
more restrictive. The first set included only dyads in which at least one
member had already entered one or more alliances. The second and most
restrictive definition of the risk set included dyads in which both members
had entered at least one other alliance. The results obtained with different
sets were comparable. The results reported here are based on the complete
risk set.

Interdependence.—This variable measures the extent to which organi-

zations may need each other to access critical resources and capabilities.
Prior research (Nohria and Garcia-Pont 1991; Shan and Hamilton 1991)
and our own extensive fieldwork suggest that interdependence between
organizations in these industries resulted primarily from the quest for
complementary capabilities and resources, but identifying and measuring
this complementarity is not an easy task. Complementarity between two
organizations can arise when (a) there is a gap between the specific capa-
bilities controlled by each organization and those they need to pursue their
strategy and (b) this gap can be filled at least partially by accessing the

10

Hallinan and Sorensen (1985) used a similar dyadic approach in examining the ef-

fects of ability groups in classrooms on the patterns of student friendships formed.
Fernandez (1991) examined the effects of informal and formal ties on leadership rela-
tions within organizations using such an approach. Both studies, however, used cross-
sectional data.

1460

Networks

capabilities controlled by the other organization while being able to offer
something of value in return. Organizational needs and capabilities, how-
ever, are multifaceted and ambiguous; assessing them across a large num-
ber of organizations poses a formidable measurement problem. In addi-
tion, an index of complementarity for all possible pairs of organizations
requires measuring the extent to which the capabilities of one organization
can “complement” the capabilities of every other organization in the in-
dustry. We therefore used several approaches to assess complementary
capabilities and resources that could create interdependence between or-
ganizations and conducted a statistical analysis to capture any unobserved
affinity between organizations not adequately accounted for with the vari-
ables included in the analyses.

The measure of interdependence reported in this article is based on two

key dimensions that drive complementarity between firms in our global
setting: national origin and industry subsegment. National origin captures
the geographical clustering of capabilities in the global economy. Regional
contexts circumscribe important sets of unique organizational capabilities
and resources, which resulted from specific historical and institutional
processes (Porter 1990; Hamilton and Biggart 1988). In addition, interde-
pendence across different geographical regions can result from the need
to gain access to markets in those regions. Organizations from different
regions are therefore more likely to have complementary capabilities and
to benefit from strategic alliances with each other (Shan and Hamilton
1991). We captured these regional differences by grouping organizations
in three categories—American, Japanese, and European—which corre-
spond to the three major global markets, as well as to three relatively
distinct historical and institutional settings.

Industry subsegment captures complementarity across different techno-

logical “niches” within an industry. Each of these niches corresponds to
clusters of firms that share specific sets of capabilities and resources. Firms
in different niches are more likely to have complementary capabilities that
can make them interdependent and lead to alliances between them (Noh-
ria and Garcia-Pont 1991). Building on this insight, we identified broad
subsegments that define distinctive clusters of organizations within each
industry. We identified two subsegments in the new materials sector (ce-
ramics and polymer composites), four in industrial automation (discrete
automation, process automation, software, and robotics), and two in the
automotive sector (automobile assemblers and suppliers).

We measure interdependence between any two organizations within an

industry as the normalized euclidean distance between those organiza-
tions, computed from the matrix that captures the national origin and
industry subsegment of each firm and computed to capture the absence
of overlap of activities. The greater the normalized euclidean distance

1461

American Journal of Sociology

between two organizations, the more likely they are to possess comple-
mentary capabilities and resources, and the higher their interdependence.
To check the validity of this measure, we performed a cluster analysis of
the euclidean-distance matrix for each industry and identified clusters of
organizations with similar national and technological profile. We identi-
fied seven distinct clusters in the new materials sector and nine clusters
each in the industrial automation and automotive sectors.

11

The composi-

tion of those clusters was then checked with recent studies of similar in-
dustries (Nohria and Garcia-Pont 1991), as well as with a classification
of the same firms in discrete strategic clusters by a panel of industry ex-
perts. The high convergence (80%) between the groups formed by the
experts and those obtained from clustering the matrix of euclidean dis-
tances between the firms validates the use of these distances as a measure
of interdependence. To further assess the similarity between the groups
obtained through cluster analysis and the continuous measure used in this
study, we constructed a dummy variable coded “1” if the organizations
in a dyad belonged to different clusters and “0” if they belonged to the
same cluster. The results obtained using this variable were similar to the
ones based on the continuous measure of interdependence.

We also tried to account for interdependence between organizations in

our sample by considering a series of firm-specific attributes—such as size
and financial performance—that capture resource availability and con-
straints that typically influence the propensity of organizations to enter
alliances. We discuss these attributes below as control variables. Finally,
we tried to capture any residual interdependence with a statistical model
that controls for unobserved factors that might affect the likelihood of
alliances between specific firms. Details of this follow in the next section.

Network measures.—To compute our network measures, we con-

structed adjacency matrices representing the relationships between the
organizations in each industry for each year. We included all alliance ac-
tivity among industry panel members for the previous five years. One
concern with such a design is left-censoring, which is an issue because
many of the sample organizations existed prior to the start of the alliance
observation period in 1981. Additional alliance data were collected for the
alliance activity of this sample of organizations for 11 years, dating back to
1970 to minimize left-censorship effects. These data confirm that alliance

11

Our partitioning of industries created clusters of organizations that are akin to the

concept of “strategic groups” (for reviews, see Thomas and Venkatraman 1988; Barney
and Hoskisson 1990; Reger and Huff 1993). Our clusters are closest to Porter’s (1979)
original definition of a strategic group as a set of organizations within an industry
that are similar to one another in one or more strategic dimensions, such as skills,
resources, goals, and historical development.

1462

Networks

activity was negligible until 1980, when there was an explosion of alliances
(Hergert and Morris 1988).

We made a number of choices in constructing these matrices about the

treatment of different types of alliances, the accumulation of multiple ties
by the same partners, and the past alliances that should be included. These
choices were all tested against alternatives to ensure the robustness of our
findings. First, alliances were weighted by their strength, as represented
by their formal governance structure, using a seven-point scale (Contrac-
tor and Lorange 1988; Nohria and Garcia-Pont 1991), and the results were
checked against a simple dichotomous measure. Second, to take into ac-
count the cumulative history of alliances between organizations, we used
a Guttman scale that captures the score of the strongest alliance formed
by the two organizations, checking the results against simple additive
scores and normalized additive scores. Third, we used a moving window
of five years of prior alliances, based on research suggesting that the nor-
mal life span for most alliances is usually no more than five years (Kogut
1988). We checked the results against networks that included all previous
alliances dating back to 1970 in the construction of the networks.

Relational embeddedness.—This construct indicates the extent to

which a pair of organizations (dyad) had direct contact with each other
in the past. For our longitudinal panel of pairs of organizations for 1981–
89, we operationalized relational embeddedness as the number of alliances
dyad members had entered with each other in the previous five years.

Structural embeddedness.—This construct indicates the extent to

which a given pair of organizations shared common partners from past
ties. For each dyad-year record, we computed the number of partners
shared by the two organizations in a dyad as a result of their alliances in
the previous five years. To differentiate structural embeddedness from
relational embeddedness, we set common ties to zero if the members of
a dyad sharing common ties had entered at least one previous direct alli-
ance with one another (cf. Mizruchi 1992, p. 126).

Positional embeddedness.—This construct indicates the extent to

which the organizations in a dyad occupy similar or different network
positions. We first computed a measure for the position of each organiza-
tion and then used those as inputs to compute dyadic values. We measured
the position of an organization in the emerging network of alliances using
the Bonacich (1987) eigenvector measure of network centrality, a choice
that is consistent with prior efforts to capture the position or role of an
organization in a relational network (Mizruchi 1993; Podolny 1994). Using
this measure, the most central organizations are those linked to many or-
ganizations, which are in turn linked to several other organizations. We
computed the eigenvector measure of the network centrality of each orga-
nization for each year and expressed the scores relative to the most central

1463

American Journal of Sociology

organization in the network for that particular year (C

max

⫽ 1). To capture

the joint centrality of the dyad, we computed the geometric mean of the
centrality scores for each member of the dyad and then normalized it by the
industry maximum (Mizruchi 1993). The larger the score, the more the two
organizations occupied a central role in the network. To capture the simi-
larity in centrality in a dyad, we computed the ratio of the smaller to the
larger centrality score of the two organizations. The closer this ratio was to
1.0, the more similar were the two organizations’ positions in the network.

Structural differentiation.—Our indicator for this construct reflects the

nature of the differentiation that characterizes the specific interorganiza-
tional systems under investigation. Our positional analysis revealed the
emergence of a center-periphery structure in all the three industry net-
works. At the system level, the emergence of such a structure is parallel
to an increase in network centralization. Thus, an indicator that captured
the extent of network centralization could adequately represent the type
of differentiation observed in our networks. For each observation year,
we measured the structural differentiation of the network as the level of
network centralization in that year. Following Wasserman and Faust
(1994), we measured network centralization as the standard deviation of
the eigenvector centrality scores of the organizations in the industry for
that year. Because we normalized eigenvector centrality scores by the
highest centrality in each industry and year, our measure captured the
relative internal differentiation of the system for each industry and in each
given observation year. The changes over time in the measure of struc-
tural differentiation captures the emergence of the center-periphery struc-
ture that characterizes the industry networks. The average standard devi-
ation of centrality across the three industries displayed a linear monotonic
increase over time, ranging from .15 in t

0

(1981) to .41 in t

8

(1989). These

aggregate figures adequately represent the pattern observed in each of the
three industries and further highlight the progressive centralization of the
networks as alliances accumulate. We tested the hypothesized moderating
influence of structural differentiation on interdependence and embed-
dedness with interaction terms that were constructed using the product-
term approach (Jaccard, Turrisi, and Wan 1990).

Control Variables.—We included as controls a number of variables

known or expected to affect the alliance activity of organizations. These are
network density, time, sector, organization-level effects, and a set of finan-
cial measures capturing organizational differences in some key resources.

An alternative interpretation for the endogenous embeddedness dy-

namic proposed here is a density-dependence argument linking the num-
ber of previous alliances to the legitimacy of this new form of business
relationship. Ecological density-dependence arguments claim that there
is an initial positive impact of density on founding rates of organizations

1464

Networks

via the effect of density on the legitimacy of the new organizational form
(Hannan and Freeman 1989, p. 132). Applied to this context, it would
suggest that the growth in alliances may be the result of a bandwagon
effect (Venkatraman, Loh, and Koh 1994). Thus, one could argue that
structural differentiation might simply be capturing the progressive legiti-
mization of alliances as a valid form of interorganizational cooperation,
rather than the informational effects proposed in our model. If this were
the case, the growth of alliances would be driven by the effects of density-
dependent legitimization rather than by the increase in the availability of
information captured by our notion of structural differentiation.

To account for this alternative explanation, we included a measure

called alliance density, defined as the cumulative number of alliances
within the industry in the previous five years, divided by the total number
of possible alliances in the system. If the effect of structural differentiation
is only capturing density-driven legitimacy, the inclusion of alliance den-
sity should make the effect of structural differentiation insignificant, thus
bringing into question the validity of our claims. The endogenous network
dynamic model, however, does not preclude a legitimization effect, be-
cause density-dependence and structural-differentiation effects need not
be mutually exclusive.

To control for unobserved temporal factors that may influence alliance

formation, we included dummy variables for each year. Such factors could
include a progressive legitimization not accounted for by the simple accu-
mulation of alliances—or unspecified events that may alter the likelihood
of new alliances. For simplicity of presentation, we then reestimated these
effects using a single variable, time, which ranges from “0” to “8” (with
the default year being 1981) and assumes linearity in the effect of time.
We observed no differences in the results based on the alternative controls
for time. We also controlled for sector differences with two dummy vari-
ables, labeled new materials and industrial automation, using the automo-
tive sector as the default sector.

Unobservable organization-level effects were captured by two variables

indicating the prior alliance experience of each partnering organization
in each dyad (Heckman and Borjas 1980). We computed a measure for
each organization that captured the total number of alliances it had previ-
ously entered in the past five years. These variables, labeled alliance his-
tory, firm 1 and firm 2, capture the possibility of repetitive momentum in
individual organizations’ alliance activities, as well as unobserved factors
affecting each organization’s proclivity to form partnerships. It is worth
noting that these measures are akin to the network analysis notion of “de-
gree centrality,” which is defined as the number of ties in which an actor
is involved (Freeman 1979).

We also included a series of financial measures to capture the differ-

1465

American Journal of Sociology

ences across the organizations in a number of key dimensions. Insofar as
differences in the control of financial resources may result in complemen-
tarities that lead to strategic alliances, these controls also can be discussed
as an additional way of capturing possible sources of interdependence
between organizations (Ghemawat, Porter, and Rawlinson 1986; Barley,
Freeman, and Hybels 1992). For each such dimension, we computed a
ratio of the smaller to the larger organization. In this way, we controlled
for relative differences in financial resources and performance that may
have influenced the likelihood of alliance formation between the two orga-
nizations. The first dimension was size, measured as total sales in the in-
dustry; the second was performance, captured as return on assets normal-
ized to the industry mean (a common measure of performance in
managerial research); the third was liquidity. Organizations frequently
enter alliances to share the costs of new projects, particularly those involv-
ing large resource outlays and risks. In this context, relative liquidity,
which reflects the short-term resources available to an organization, is
important. We used the “quick ratio”—defined as current assets minus
inventory, divided by current liabilities—to measure liquidity (Dooley
1969). Last, we examined solvency differences across the two organiza-
tions in each dyad. We used an organization’s relative-debt profile within
its industry, measured as the total amount of long-term debt divided by
the organization’s current assets.

We also examined whether each of the above financial attributes for

each organization in the dyad separately influenced alliance formation.
Thus, for each organization in a dyad, we introduced separate variables
indicating the size, performance, liquidity, and solvency of each organiza-
tion (eight variables). These variables had no effect on our main results and
so, in the interest of parsimony, we omitted them from our final analysis.

Table 1 lists the variables, their summary definitions, and their pre-

dicted effect on the probability of alliance formation, while table 2 dis-
plays descriptive statistics and a correlation matrix for all the variables in
the analysis.

Model

We modeled alliance formation using the following dynamic panel model,
in which a variable’s positive coefficients indicate that it promotes alliance
formation:

12

12

It is important to note that this approach is distinct from that using the class of

models known as network effects or endogenous feedback, which is familiar to net-
work analysts (Marsden and Friedkin 1993). The postulated network effects here re-
sult from a lagged network of cumulative prior ties until the previous year, rather
than being linked to network elements in the same period.

1466

TABLE 1

Definitions and Predicted Signs of Variables

Variable

Definition

Prediction

Alliance ............................................

Whether two firms formed an alliance

Dependent

in a given year

variable

Interdependence .............................

Normalized euclidean distance score

⫹

capturing the absence of overlap of
activities between firms

Structural differentiation ...............

SD of the normalized prominence

⫹

scores among firms in the industry

Repeated ties ...................................

Number of prior alliances between

⫹

the firms in prior five years

Common ties ...................................

Number of common partners shared

⫹

by previously unconnected firms in
prior five years

Joint centrality ................................

Geometric mean of multiple of cen-

⫹

trality of both firms normalized by
industry maximum

Centrality ratio ...............................

Ratio of centrality of lesser to greater

⫹

value

Network density .............................

Ratio of cumulative alliances in an in-

⫹

dustry divided by the number of
possible alliances

Time .................................................

Year value for each record, ranging

NP

from zero to eight

New materials ................................

“1” if firms are in the new materials

NP

sector (default: automotive)

Industrial automation ....................

“1” if firms are in the industrial auto-

NP

mation sector (default: automotive)

Alliance history, firm 1 ..................

Number of prior alliances entered by

NP

firm 1 in the dyad

Alliance history, firm 2 ..................

Number of prior alliances entered by

NP

firm 2 in the dyad

Size ...................................................

Ratio of sales of smaller to larger

NP

partner

Performance ....................................

Ratio of performance (ROA) of lesser

NP

to greater firm value

Liquidity ..........................................

Ratio of liquidity (quick ratio) of

NP

lesser to greater firm value

Solvency ..........................................

Ratio of solvency (long-term debt) of

NP

lesser to greater firm value

Note.—NP

⫽ no prediction.

TABLE

2

Descriptive

Statistics

and

Correlation

Matrix

V

a

ri

a

b

le

s

M

e

a

n

S

D

L

o

w

H

ig

h

12

3456

789

1

0

1

1

1

2

1

3

1

4

1

5

1

6

1

7

Alliance

.................................

.21

.09

0

1

.00

⋅⋅

⋅

Interdependence

..................

1.42

.28

0

2.47

.16

Structural

d

ifferentiation

....

.29

.08

.15

.41

.24

.00

Repeated

ties

........................

.12

.40

0

5

.00

.19

.10

.36

Common

ties

........................

.56

1

.21

0

9.00

.15

.03

.19

.32

Joint

centrality

.....................

.21

.32

0

1

.00

.16

.07

.11

.25

.38

Centrality

ratio

....................

.46

.21

0

1

.00

.21

.01

⫺

.02

.12

⫺

.06

.26

Network

d

ensity

..................

.16

.09

.06

.22

.16

⫺

.04

.43

.30

.21

.08

.04

Time

......................................

5.72

1.98

0

8

.00

.02

.11

.38

.06

.08

.13

⫺

.11

.49

New

materials

......................

.44

.49

0

1

.00

.00

.04

.19

⫺

.05

⫺

.12

⫺

.10

.03

.21

.00

Industrial

automation

.........

.23

.42

0

1

.00

⫺

.01

.01

⫺

.22

⫺

.03

⫺

.08

⫺

.14

⫺

.13

.10

.00

⫺

.24

Alliance

h

istory,

fi

rm

1

.......

3.00

2.79

0

1

6

.12

.09

.07

.21

.32

.35

.00

.03

.32

.03

⫺

.06

Alliance

h

istory,

fi

rm

2

.......

3.13

2.90

0

1

6

.10

.13

.12

.18

.27

.42

.01

.18

.35

⫺

.05

⫺

.13

.14

Size

........................................

.27

.25

.09

.94

.02

⫺

.21

.01

.05

.06

⫺

.03

.24

.02

.01

⫺

.03

.00

⫺

.01

⫺

.05

Peformance

...........................

.35

.29

.13

.90

.00

⫺

.09

⫺

.03

.00

.00

.00

.13

.11

⫺

.01

.00

⫺

.01

.00

.00

.00

Liquidity

...............................

.24

.19

.07

.88

.01

⫺

.02

.00

.02

.03

.00

.07

⫺

.06

.04

.00

.013

.03

.00

.38

.17

Solvency

...............................

.64

.22

.02

.80

.00

⫺

.05

.09

.01

.03

.00

.16

.00

⫺

.03

.05

⫺

.05

.04

.05

.05

.00

.08

⋅⋅⋅

Networks

p

ij

(t)

⫽

Φ

(a

⫹ bx

ij

⫹ cy

ij

(t

⫺ 1) ⫹ u

ij

),

where p

ij

(t) is the probability at time (t) of the announcement of an alliance

between organizations i and j; x

ij

is a time-constant vector of covariates

characterizing organizations i and j; y

ij

(t

⫺ 1) is a time-varying vector

of covariates characterizing organizations i and j; u

ij

is unobserved time-

constant effects not captured by the independent variables;

Φ

is the nor-

mal cumulative distribution function.

We employed a random-effects panel probit model that accounts for

unobserved heterogeneity and was implemented here using LIMDEP 6.0
(Butler and Moffitt 1982). Details about our choice of model and the neces-
sity for accounting for unobserved heterogeneity are provided in the ap-
pendix.

One concern with analyzing dyadic data is possible interdependence

across observations (Lincoln 1984; Mizruchi 1989). To ensure the ro-
bustness of our results, we employed a procedure similar to the Multivari-
ate Regression Quadratic Assignment Procedure (MRQAP), routinely
used by researchers studying dyads (Krackardt 1987, 1988; Manley 1992;
Mizruchi 1992). The percentage of frequency with which the results in
the random-sample simulations exceeded the original estimates was far
less than 5% in all instances, which attests to the robustness of our probit
estimates. Details of these tests are reported in the appendix.

RESULTS

Table 3 presents probit estimates for the effects of factors influencing the
formation of new ties between organizations. The coefficients indicate
how a change in an independent variable in the previous year affects the
probability of two organizations forming a new alliance during the current
year.

Model 1 presents a baseline containing an array of control variables.

These include the density of alliances in the sector, time, dummy variables
for industrial sectors, and controls for each organization’s previous alli-
ance experience (labeled alliance history) as well as their similarity in a
series of financial indicators. The density of alliances in the prior time
period has a positive impact on new alliance formation, which suggests
possible legitimization effects. The introduction of alliance density in the
model makes the effect of time nonsignificant, suggesting that most linear
time-related factors are captured by cumulative industry density. There
was a significant improvement in the chi-square statistic once we intro-
duced alliance density, which further suggests that the density of the net-
work may mediate the influence of time on alliance formation. In separate

1469

TABLE

3

Random-Effects

Panel

Probit

Estimates

Variable

1

23456789

1

0

Constant

................................................................................

1.33*

1.09*

.58

.52

.50

.52

.52

.43

.37

.39

(.24)

(.21)

(.49)

(.48)

(.48)

(.47)

(.47)

(.34)

(.35)

(.37)

Interdependence

...................................................................

⋅⋅

⋅

1.15*

.91*

.69*

.66*

.59*

.57*

.52*

.56*

.58*

(.39)

(.30)

(.21)

(.20)

(.17)

(.17)

(.16)

(.17)

(.18)

Structural

differentiation

....................................................

3.78*

3.51*

3.03*

2.93*

2.88*

2.46*

2.39*

2.08*

(1.02)

(1.01)

(.91)

(.88)

(.89)

(.74)

(.70)

(.63)

Repeated

ties

........................................................................

1.38*

.73*

.61*

.60*

.43*

.48*

.44*

(.20)

(.15)

(.14)

(.14)

(.12)

(.13)

(.12)

Common

ties

........................................................................

1.26*

1.18*

1.03*

.64*

.94*

.98*

(.14)

(.20)

(.21)

(.18)

(.20)

(.22)

Joint

centrality

.....................................................................

.69*

.48*

.42*

.29*

.38*

(.10)

(.13)

(.13)

(.07)

(.13)

Centrality

ratio

.....................................................................

.08

.06

.07

.18

(.05)

(.05)

(.05)

(.10)

Structural

differentiation

⫻

interdependence

..................

⫺

1.79*

⋅⋅

⋅

(.40)

Structural

differentiation

⫻

joint

centrality

.....................

1.38*

⋅⋅

⋅

(.22)

Structural

differentiation

⫻

centrality

ratio

.....................

⋅⋅

⋅

1.08*

(.20)

1470

Network

density

..................................................................

.16*

.15*

.11

.10

.11

.11

.10

.09

.12

.12

(.03)

(.03)

(.06)

(.06)

(.06)

(.07)

(.07)

(.07)

(.07)

(.08)

Time

......................................................................................

.07

.06

.03

.03

.03

.00

.00

.01

.01

.03

(.05)

(.05)

(.05)

(.04)

(.04)

(.04)

(.05)

(.03)

(.03)

(.02)

New

m

aterials

......................................................................

⫺

.23

⫺

.23

⫺

.20

⫺

.21

⫺

.19

⫺

.16

⫺

.15

⫺

.16

⫺

.15

⫺

.17

(.15)

(.14)

(.14)

(.14)

(.14)

(.11)

(.12)

(.10)

(.10)

(.12)

Industrial

automation

..........................................................

.03

.06

.05

.05

.04

.02

.06

.06

.02

.04

(.02)

(.05)

(.04)

(.05)

(.05)

(.05)

(.05)

(.04)

(.04)

(.06)

Alliance

history,

firm

1

.......................................................

.42

.31

.29

.23

.18

.15

.10

.10

.12

.08

(.27)

(.23)

(.23)

(.21)

(.20)

(.21)

(.16)

(.16)

(.15)

(.14)

Alliance

history,

firm

2

.......................................................

.11

.24

.19

.18

.17

.14

.17

.16

.15

.16

(.07)

(.15)

(.14)

(.16)

(.16)

(.16)

(.16)

(.16)

(.16)

(.15)

Size

........................................................................................

⫺

.52*

⫺

.47*

⫺

.39*

⫺

.35*

⫺

.32*

⫺

.31*

⫺

.25*

⫺

.20*

⫺

.18*

⫺

.23*

(.12)

(.12)

(.11)

(.11)

(.10)

(.10)

(.05)

(.05)

(.05)

(.06)

Performance

..........................................................................

⫺

.15*

⫺

.09

⫺

.05

⫺

.03

⫺

.03

⫺

.00

⫺

.02

⫺

.02

⫺

.03

⫺

.05

(.03)

(.05)

(.04)

(.04)

(.04)

(.04)

(.04)

(.04)

(.04)

(.04)

Liquidity

...............................................................................

⫺

.34

⫺

.30

⫺

.27

⫺

.26

⫺

.20

⫺

.18

⫺

.17

⫺

.21

⫺

.19

⫺

.16

(.19)

(.18)

(.19)

(.18)

(.14)

(.13)

(.11)

(.10)

(.10)

(.12)

Solvency

................................................................................

.10

.08

.07

.06

.05

.06

.05

.05

.06

.05

(.12)

(.11)

(.10)

(.11)

(.11)

(.11)

(.10)

(.10)

(.10)

(.10)

Rho

........................................................................................

.44

.43

.36

.35

.32

.30

.29

.24

.21

.23

(.13)

(.13)

(.10)

(.10)

(.09)

(.09)

(.09)

(.07)

(.07)

(.07)

N

............................................................................................

7266

7266

7266

7266

7266

7266

7266

7266

726

6

7266

Chi

square

............................................................................

45.73*

49.34*

58.47*

65.62*

69.76*

74.57*

76.71*

83.42*

87.11*

85.04*

N

ote.

—

S

Ds

are

in

parentheses.

*

P

⬍

.05.

1471

American Journal of Sociology

analyses, we also introduced a variable capturing the number of alliances
announced in the industry in the previous year, but this variable was
not significant once we controlled for industry density and thus was not
included in the models.

The variables for alliance history of each organization were insignifi-

cant across all models, indicating that the individual prior experience
with alliances of each organization within a dyad did not make an alliance
between them more likely. Model 1 also included ratios measuring the
similarity between pairs of organizations in size, performance, liquidity,
and solvency. Except for size, similarity of financial indicators did not
have a significant impact on the probability of alliance formation. Alli-
ances are more likely to occur between organizations of different size.
While these ratios are introduced as controls, they also capture any resid-
ual effects of interdependence not accounted for by our measure of inter-
dependence. We also examined whether the above financial attributes
introduced separately for each organization in the dyad influenced
alliance formation, but these variables had no impact on our main re-
sults and were omitted from our final analysis in the interest of parsi-
mony.

Model 2 introduced our measure of interdependence between the mem-

bers of the dyad. As predicted in hypothesis 1, organizations separated
by a larger normalized euclidean distance in the matrix that captures in-
terorganizational interdependence were more likely to enter alliances.
This result is congruent with research on the role of interdependence in
alliance formation and also helps enhance the construct validity of our in-
dicator. Our alternative measure for interdependence, using membership
in the clusters corroborated by industry experts, yielded similar results.

We introduced structural differentiation in model 3. As predicted by

hypothesis 6, structural differentiation has a strong positive effect on alli-
ance formation. Introducing this measure also leads to a significant im-
provement in the fit of the model, as measured by the chi-square statistics.
Moreover, the effect of density became nonsignificant at the .05 level once
we introduced structural differentiation into this model. This suggests that
the systemic effects on tie formation captured by density may actually
be mediated by the structural differentiation of the network. Although
hypothesis 6 was not formulated as an alternative for a density-dependent
legitimization process, the statistical insignificance of the density effect in
model 3 suggests that the increase in the probability of alliance formation
is perhaps best explained by the growing differentiation of the network
structure. Thus, the upward-sloping rate of alliance formation during the
1980s may be prompted by the emergence of a differentiated social struc-
ture that made it easier for organizations to identify suitable partners in

1472

Networks

an uncertain environment, rather than a consequence of a legitimization
effect driven by the mere accumulation of ties over time.

13

Models 4–7 test the effect of the various embeddedness mechanisms on

alliance formation, as predicted by hypotheses 2–5. Models 4 and 5 con-
firm the expected influence of both relational and structural embed-
dedness on subsequent alliance formation, as proposed in hypotheses 2
and 3. In model 4, the positive and significant coefficient of repeated ties
indicates that the presence of prior ties between organizations in the previ-
ous five years positively influences the likelihood of their forming a new
alliance. In model 5, the positive and significant effect of common ties
indicates that shared third-party ties between previously unconnected or-
ganizations increases their probability of entering an alliance. The effects
of both direct and indirect ties on alliance formation remain significant
across all the models.

14

Models 6 and 7 examine the role of the position of organizations in the

emergent structure of interorganizational ties on their alliance behavior.
Model 6 shows that the probability of two organizations forming a new
alliance increases with the joint centrality of the potential partners, as
predicted by hypothesis 4. The evidence for hypothesis 5, which predicts
an increase in the probability of an alliance between organizations with
similar centrality, is less conclusive. The results indicate that the differ-
ence in centrality scores does not have a statistically significant influence
on the likelihood of an alliance between two organizations. In separate

13

We also assessed the impact of diversity-dependent legitimation on alliance forma-

tion by modeling the effects of the diversity of alliances formed within industries.
Diversity was conceptualized in terms of the kinds of governance structure organiza-
tions used to formalize their alliances. We assessed diversity with two sets of measures.
First, we computed the reciprocal of the Herfindahl index for governance structure
of prior alliances. Second, we used a specification akin to Blau’s index of heterogeneity
(Blau 1977; Powell et al. 1996). We computed the proportion of organization i’s ties
of type j until year t, out of the total number of ties the organization had entered until
that year, denoted as P

it, j

. We defined six types of governance structures. We computed

the index of diversity Y

it

by subtracting the summation over all j of the square of P

it , j

that is, Y

it

⫽ 1 ⫺

∑

j

(P

it , j

)

2

and (1

ⱕ j ⱕ 6). The results were insignificant for both

measures of diversity across all models and were thus not included with the final
analyses.

14

We also tested polynomial transformations of the two cohesion variables to account

for nonlinear effects. The results suggest that the relationship between previous alli-
ances and future alliances within the dyads is best described as an inverted

U

-shaped

relationship, captured by a second-order polynomial function. The effect, however,
is exponential for shared common ties between unconnected organizations. As the
number of common ties between organizations increases, the likelihood of their allying
with each other increases disproportionately. The inclusion of the polynomial transfor-
mations does not affect the results obtained with the linear forms. We report the results
of the linear model for the sake of parsimony.

1473

American Journal of Sociology

estimations, however, we found that the ratio of centrality was positive
and significant if joint centrality was excluded from the model. To inter-
pret this result, it is worth noting that the ratio of centrality approaches 1.0
for any two organizations that have similar centrality scores, regardless
of their absolute level of centrality. In other words, while two “central”
organizations are similar, so are two “peripheral” ones. Yet our positional
analysis of the networks has shown that peripheral organizations are
much less likely to enter alliances. If they do, they are likely to do so with
central organizations, not with other peripheral ones. Central organiza-
tions form ties with other central organizations and, to a lesser extent,
with less central organizations. Thus, the homophily tendency implicit in
hypothesis 5 only applies to central organizations. Once the joint centrality
of the dyad is controlled for, the effect of homophily is no longer signifi-
cant. Viewed in this light, these results are consistent with the distinction
introduced by Mizruchi (1993) between “central” and “peripheral” role
equivalence, which suggests that there are significant differences in the
behavior of these two types of actors that cannot be easily captured by a
single sociological construct.

Models 8, 9, and 10 assess the moderating influence of structural differ-

entiation on both endogenous and exogenous drivers of alliance formation.
We had predicted that structural differentiation in the system would mod-
erate the influence of positional embeddedness on alliance formation (hy-
pothesis 8). This prediction should translate into a significant and positive
coefficient for the interaction between structural differentiation and both
joint centrality and similarity in centrality. But we also predicted that the
effect of exogenous interdependence on alliance formation would diminish
with structural differentiation (hypothesis 7). This effect should yield a
significant and negative coefficient for the interaction between interdepen-
dence and structural differentiation. We tested these models separately
because of concerns of multicollinearity across the interaction terms.

Model 8 introduces an interaction term between interdependence and

structural differentiation. The negative coefficient for the interaction term
supports hypothesis 7 and suggests that the explanatory power of inter-
dependence diminishes with the growing differentiation of the social
structure of interorganizational ties, but interdependence on its own has
a positive impact on alliance formation across all models. Thus, while
exogenous factors do influence the creation of new ties, the increasing
structural differentiation of the network enables organizations to use this
network as a source of information for their future partnerships, which
mitigates the effects of exogenous interdependence on the formation of
new alliances.

Models 9 and 10 introduce the interactions between structural differen-

tiation and the two indicators of positional embeddedness. Model 9 tests

1474

Networks

the interaction between structural differentiation and joint-dyad cen-
trality, and model 10 adds the interaction between structural differentia-
tion and the similarity in centrality within the dyad. Together, these two
models test the contingent influence of positional embeddedness on alli-
ance formation (hypothesis 8). Both models show significant positive ef-
fects for the interaction between positional embeddedness and structural
differentiation. The impact of both joint centrality and similarity in cen-
trality increases with the level of structural differentiation of the emerging
network. This suggests that the effect of positional embeddedness on alli-
ance formation increases with the level of structural differentiation of the
emerging network. As predicted by our framework, the effective impact
of positional-embeddedness mechanisms on subsequent tie formation is
thus contingent upon the level of structural differentiation of the network.

Although similarity in centrality was not a significant predictor of alli-

ance formation, the interaction between this variable and structural differ-
entiation is statistically significant. This suggests that, with the growth of
structural differentiation, organizations may become increasingly aware
of differences in centrality when choosing a partner, although this ten-
dency is not strong enough to make the difference in centrality statistically
significant during the period of observation. The more the respective posi-
tions of organizations in the network become apparent, the more difficult
it may become for a peripheral organization to build alliances with a cen-
tral one. Although this does not mean that such alliances will not occur,
it does suggest that peripheral organizations may need to possess some
unique attributes that can enhance their attractiveness as alliance partners
with central organizations.

DISCUSSION

The central message of this research is that the formation of a new interor-
ganizational network structure results from a longitudinal dynamic in
which action and structure are closely intertwined. Our model portrays
the social structure of interorganizational relations as a “macro” phenome-
non emerging out of the “micro” decisions of organizations seeking to gain
access to resources and to minimize the uncertainty associated with choos-
ing alliance partners. The network structure that results from the accumu-
lation of those ties increasingly becomes a repository of information on
potential partners, helping organizations decide with whom to form new
alliances. The emerging alliance network consequently increasingly in-
fluences organizational action. In this model, the dialectic between macro
and micro is thus translated into the dialectic between structure and ac-
tion.

The results show that both interdependence and network embed-

1475

American Journal of Sociology

dedness factors have a significant impact on new alliance formation. Con-
sistent with prior research, organizations build ties with other organiza-
tions that have complementary resources and capabilities, but they also
take into consideration the position the potential partners have in the
emerging social structure of the network. The influence of interdepen-
dence and network factors is contingent on the level of structural differen-
tiation of the social system. The role of positional embeddedness in alli-
ance formation increases with the growing structural differentiation of
the emerging interorganizational network, while the impact of exogenous
factors declines. These findings support our claim that the increasing dif-
ferentiation of social structure reflects a process by which the network
becomes a repository of information about potential partners. The higher
the structural differentiation of the emerging network, the more organiza-
tional decisions about new partnerships are guided by endogenous net-
work considerations and the less by exogenous factors.

As we interpret the results, the emerging alliance network progressively

internalizes relevant information about competencies, needs, and reliabil-
ity of potential partners. The embeddedness mechanisms enable organiza-
tions to identify complementary and reliable partners, reducing the haz-
ards of cooperation. This interpretation implicitly assumes that there is
no tension between instrumental and social drivers of alliance formation,
yet our results do not preclude the possibility of such a tension. The emer-
gence of a network structure increases the information available to organi-
zations, but it may also limit the effective range of potential partners orga-
nizations are likely to consider. The possibility that instrumental
rationality could be subordinated to embedded action has been empha-
sized often in neoclassical economics. Yet sociologists have also suggested
that in some situations, social structures may actually hinder, rather than
help, the pursuit of economic interest. Studies of ethnic entrepreneurs
(Portes and Sensenbrenner 1993) and middle managers (Gargiulo and Be-
nassi 1998) suggest that the same social mechanisms that facilitate instru-
mental cooperation may also have a “dark side” that can damage an
actor’s ability to pursue instrumental goals (Gulati and Westphal 1999).
One of the themes in these studies is that membership in cohesive clus-
ters hinders the actor’s ability to build cooperative ties with actors not
connected to that cluster. A similar risk is implicit in our structural-
embeddedness mechanism and could equally limit the formation of ties
within a cohesive “core” of central organizations.

By relying on an evolving social structure, boundedly rational organiza-

tions effectively diminish the uncertainty associated with picking part-
ners, but gains in partner reliability may be offset by the limitations on
the choice of potential partners. Some features of strategic alliances sug-
gest that this trade-off may be more than a theoretical possibility. The

1476

Networks

hazards of interorganizational cooperation, coupled with the difficulty of
assessing complementary capabilities and the often ambiguous link be-
tween alliances and organizational performance, may prompt organiza-
tions to enter “secure” partnerships that could fail to realize the full poten-
tial of strategic alliances they could have entered. Future research in this
field should investigate a possible trade-off between the reduction of un-
certainty attained through embedded partnerships and pursuit of the in-
strumental logic that promotes interorganizational cooperation. If such a
trade-off exists, the search for an alliance partner could result in a path-
dependent process (Arthur 1989), in which instrumental rationality is at
times subordinated to considerations of embeddedness. While available
evidence suggests that organizations usually avoid the perils of excessive
involvement with the same partner (Gulati and Lawrence 1999), they may
still be victims of subtler forms of “overembeddedness” that could limit
their search for partners, depriving them of the full benefits of strategic
alliances. Consistent with Gargiulo and Benassi’s (1998) work on mana-
gerial networks, we could expect that organizations tied to a cohesive clus-
ter of alliance partners might run a higher risk of falling into a path-
dependent process that effectively limits their range of potential partners.

Our analysis of the structure of alliance networks uncovered the emer-

gence of core-periphery structures in the automotive, new materials,
and—to a lesser extent—industrial automation industries. The structural-
differentiation process that is at the core of our model of network forma-
tion, however, can be compatible also with alternative structural configu-
rations. In some circumstances, structural differentiation may result in a
structure with two or more blocs of organizations (or “factions”) with rela-
tively few ties across blocs. Such structures typically arise when there are
strong exogenous barriers to the formation of ties between organizations
from different blocs, like those between defense firms belonging to the
Soviet and the western blocs during the cold war. It is worth noting that
core-periphery structures are compatible with polarization, a clear exam-
ple of which is the world system before the collapse of the former Soviet
Union (see Van Rossem [1996] for analysis and discussion). In such struc-
tures, the process of structural differentiation leading to core-periphery
structures may still operate within clusters, while exogenous forces restrict
the formation of ties across clusters. The increasing globalization of the
economy, however, makes it imperative for most large business organiza-
tions to have access to all major markets and to all possible sources of
innovation. Insofar as strategic alliances are a crucial tool to attain these
goals, faction-type interorganizational structures should be less likely to
occur, except when geographical distance, competitive network dynamics,
or geopolitical considerations restrict access to other organizations that
may have complementary capabilities.

1477

American Journal of Sociology

While our model proposes that the emergence of network structures

is driven by the endogenous differentiation of this structure, emerging
interorganizational networks may not always evolve into a definite struc-
tural pattern. This could occur in certain new, extremely dynamic,
innovation-driven industries, where all players could benefit from alli-
ances with almost any other player. In the absence of players that can
establish their dominance through their superior command of financial
or other resources, like Microsoft in the software industry and the large
pharmaceutical companies in biotechnology, the evolution of the emerging
network may not reveal any definite pattern. According to our theory,
organizations in such industries would face extremely high levels of uncer-
tainty at the time of building cooperative partnerships because they would
lack the guidance of embeddedness mechanisms. In those cases, however,
it is conceivable that other networks—such as those resulting from the
circulation of engineers between firms in Silicon Valley or from the rela-
tionships between university researchers and industry researchers as in
biotechnology—may provide an alternative to the network of prior alli-
ances as a source of information tapped by organizations. While this may
result in an interorganizational network structure that partially mirrors
the pattern of the networks that provided information to organizational
decision makers, this may not necessarily occur since such alternative net-
works may have resulted from exchange processes that differ from the
ones that drive strategic alliances.

Another important issue raised by the results pertains to the relation-

ship between structural differentiation and information, on one hand, and
alliance formation, on the other. Although our model implicitly assumed
a linear relationship between the amount of information internalized in
the network structure and the differentiation of this structure, we have
also acknowledged that further increases of structural differentiation be-
yond a certain critical level could actually decrease the level of informa-
tion in the system, which in turn would have a negative effect on the
formation of new alliances. This outcome would be consistent with the
logic underlying our theory but not with the monotonic relationship be-
tween structural differentiation and alliance formation predicted by our
hypotheses. Yet the monotonic relationship in the hypotheses is a simpli-
fication warranted by our data, which cover only a segment in the devel-
opment of the observed interorganizational networks. A more extended
observation period would have allowed us to explore the relationship be-
tween structural differentiation and alliance formation in more detail.

From a theoretical standpoint, the relationship between structural dif-

ferentiation and alliance formation could take two different forms. Each
of these forms corresponds to alternative effects of new ties on the differ-
entiation process in the network structure. First, new ties may prompt a

1478

Networks

continuous increase in the differentiation of the emerging social structure.
This continuous increase would eventually result in lower levels of infor-
mation, which should reduce the probability of new alliances. The rela-
tionship between structural differentiation and new tie formation would
then correspond to an inverse

U

-shape. Second, the emerging social struc-

ture may reach a state where the creation of new ties simply reproduces a
stable pattern of distinct positions occupied by organizations with similar
relational profiles, without further increasing the differentiation of the net-
work. Such an evolution would result in stable levels of structural differ-
entiation over time, which would effectively turn it into a constant in our
dynamic models. This is consistent with existing research showing that
mature interorganizational structures typically evolve into stable positions
occupied by actors with similar network profiles (White 1981; Burt 1988).
It is unfortunate that we could not test these alternatives with our data,
since alliance networks were far from stabilizing at the end of our observa-
tion period. The evolution of structural differentiation in emerging net-
works may also be specific to the system—or type of system—under con-
sideration. In this case, future research should investigate the factors that
might affect the growth of structural differentiation as well as those that
might influence its stabilization in some mature structures.

While this article has focused on the effects of structural differentiation

on alliance formation, our construct may be important also in other areas
of research on the effects of social structures on behavior. The central
tenet of network research is that the pattern of social ties among actors
is the main driving force behind those actors’ attitudes and behaviors
(Wellman 1988). Network scholars have shown how this approach can
provide new insights into a varied set of social phenomena, including dif-
fusion of innovations (Burt 1988; Westphal, Gulati, and Shortell 1997),
social influence (Galaskiewicz and Burt 1991), political contributions (Mi-
zruchi 1989, 1992), control strategies (Gargiulo 1993), and organizational
performance in competitive situations (Burt 1992). Most network re-
search, however, assumes a relatively stable structure that creates con-
straints and opportunities for individual behavior. Our research suggests
that the effects of network structures on behavior may be contingent on
the level of structural differentiation of that network. Future longitudinal
studies on network effects on the behavior of organizations should take
into consideration not only the mechanisms through which the network
structure affects behavior but also how the level of structural differentia-
tion of this network moderates the effective impact of those mechanisms
on organizational action.

Our focus on the origin and evolution of networks complements recent

efforts to develop mathematical models of longitudinal network data, in
which the dynamics of social networks are similarly modeled as a function

1479

American Journal of Sociology

of exogenous factors and endogenous network parameters (e.g., Iacobucci
and Wasermann 1988; Carley 1990, 1991; Zeggelink 1994; Leenders 1995,
1996; Snijders 1996). These models often include specific network mecha-
nisms such as reciprocity and transitivity to spell out how previous ties
pattern the formation of future ties (e.g., Leenders 1995), which are akin to
our relational- and structural-embeddedness mechanisms. In most cases,
however, the main goal of this work has been the development of mathe-
matical models to analyze longitudinal network data, rather than to de-
velop specific theory on the factors driving the dynamics of networks. Our
research contributes to these attempts by showing empirically how the
formation of cooperative interorganizational networks results from a dy-
namic process in which networks are both a driving force and a product
of this process.

Although this article has focused on the emergence of cooperative inter-

organizational ties that take transactions out of the market logic, it never-
theless has implications that are pertinent for the development of the eco-
nomic sociology of markets. Sociologists have demonstrated that under
conditions of uncertainty and imperfect information, market players use
the network of interorganizational relationships to guide their action. The
reliance on existing networks leads to a self-reproducing market schedule
(White 1981; Leifer and White 1988). Our results suggest that the social
mechanisms that sustain a mature market structure might also play an
important role during the formation of that structure. There may be an
important difference in network dynamics between the formative and ma-
ture stages of a market structure that does not originate from differences
in the nature of the mechanisms that guide the behavior of organizations
but results from differences in the effective impact of these mechanisms
on organizational action. If the emergence of market structures follows a
pattern similar to the one we uncovered in alliance networks, the informa-
tion content of the social structure of the market is likely to be scant in
the early stages of the market formation process. This primitive market
structure would provide little guidance to market players, who should face
considerable levels of uncertainty. In a mature market, the informational
content of this network stabilizes, resulting in the markets that White
(1981, p. 518) described as “self-reproducing social structures” among or-
ganizations that evolve different roles by observing each other’s behavior.
In such markets, organizations may come and go, but the overall structure
of market transactions remains stable, as Burt (1988) has demonstrated
in his longitudinal analysis of American markets.

Our focus on the coevolution of organizational actions and of macro-

structures resulting from the cumulated networks complements recent at-
tempts to understand the coevolution of organizations and institutions
(e.g., Davis, Diekmann, and Tinsley 1994; Haveman and Rao 1997; Davis

1480

Networks

and Greve 1997). While we have focused on interorganizational networks
of cooperation, our approach could be extended to consider the potential
role of other types of interorganizational networks in enabling the creation
of new ties and in facilitating the production of institutions that may ulti-
mately regulate the subsequent production of such ties (e.g., Stern 1979).
Thus, the development of coevolutionary accounts of embedded organiza-
tional action and of the macrostructures that result from that action re-
mains an important line of inquiry that could benefit from attention to
the endogenous network factors described in this article.

This study opens several important avenues for future research. First,

scholars could examine the relative importance of endogenous embed-
dedness dynamics across a wider array of industries. Since endogenous
embeddedness is a way to cope with the hazards of cooperation, its role
on alliance formation may be affected by industry-specific factors such as
the level of technological uncertainty and the rate of change. In a prelimi-
nary examination of this question, we found that structural differentiation
played a more significant role in the high-uncertainty new materials sector
than in the more certain automotive sector. Additional research in this
direction may shed light on the contingent effects of structural differentia-
tion across industries.

Second, since structural differentiation facilitates the selection of ade-

quate alliance partners, embedded alliances formed in more differentiated
social structures should have comparatively higher levels of success. Test-
ing this proposition, however, would require detailed survey data on the
quality, duration, and relative performance of the cooperation within the
alliances (Zaheer, McEvily, and Perrone 1998). Third, since different types
of alliances entail different levels of risk, future studies could examine the
role of network embeddedness across various types of alliances.

Finally, the domain of inquiry could be expanded beyond the formation

of alliances and consider additional organizational activities such as merg-
ers and acquisitions, which also may be influenced by endogenous embed-
dedness. While each of these possibilities opens important avenues for fu-
ture research, examining them would require significant additional data.
We hope that this study will stimulate scholars to collect such data and
expand our understanding of the dynamics associated with various types
of interorganizational networks.

This article proposed a model in which the formation of interorganiza-

tional networks is the evolutionary outcome of socially embedded organi-
zational action. Our model provides a systematic link between the social
structure of an organizational field—understood in network terms—and
the behavior of organizations within the field. This link is bidirectional.
On the one hand, the emerging social structure progressively shapes orga-
nizational decisions about whether and with whom to create new ties. On

1481

American Journal of Sociology

the other hand, this social structure is produced by the (structurally
shaped) decisions of individual organizations to establish relations with
one another. Seeking an answer to the question in our title, we have shown
that interorganizational networks result not only from exogenous drivers
such as interdependence but also from an endogenous evolutionary dy-
namic triggered by the very way in which organizations select potential
partners. In this model, actors not only react to conditions of their own
making but in the process reproduce and change those very conditions.
The endogenous embeddedness model opens the way to more detailed
studies of network-formation processes that go beyond the role of exoge-
nous factors and consider the dialectic between action and structure that
is at the core of many social processes.

APPENDIX

Interdependence of Observations

We conducted a number of additional tests to address concerns of interde-
pendence across observations resulting from our dyadic approach, which
led to the presence of the same organization across multiple dyads. We
employed a procedure similar to the Multivariate Regression Quadratic
Assignment Procedure (MRQAP), routinely used by researchers studying
dyads (Krackardt 1987, 1988; Manley 1992). Our approach differs from
MRQAP in that we used the random-effects probit model instead of ordi-
nary least squares regression for each iteration of the simulation. As a
result, we randomized the key network variables for each time period for
each industry. We ran 500 iterations of a completely specified random-
effects model with a new randomized independent network variable ob-
tained by random permutations of the rows and columns in each alliance
matrix for each industry and year. The coefficients obtained were com-
pared with those obtained in the original formulation. The percentage of
frequency with which the independent variables exceeded their original
values divided by the number of permutations plus 1 (in this case, we
used 501) indicates the statistical reliability (pseudo t-test) of the original
results.

15

This test can be interpreted like conventional tests of signifi-

cance: a result of less than 5% (or, even better, 1%) provides evidence that
the original estimates are indeed accurate. The benefit of a randomization
procedure is that obtaining satisfactory results does not require an as-
sumption of independent observations, a random sample, or a specified

15

A more complete specification of this test would have entailed randomly extracting

the 500 permutations from all possible ones for each industry (Mizruchi 1992), which
was not feasible here due to the extremely large number of permutations that would
be necessary for each industry and for each year.

1482

Networks

distribution function. This procedure allowed us to assess the efficiency of
our results, a primary concern resulting from any dyadic interdependence.

The manner in which we specified our network-embeddedness effects

makes our model akin to the P* logit models recently proposed by Wasser-
man and Pattison (1996). Building on the pioneering work by Holland
and Leinhardt (1970) and on Strauss and Ikeda (1990), P* models produce
pseudo–maximum-likelihood estimators of the probability of observing a
binary tie x

ij

, conditional on the rest of the data, without having to make

the implausible assumption that the observations (dyads) are independent.
Specifically, these models build into a logistic regression parameters that
capture possible sources of interdependence of the observed dyads—such
as reciprocity, transitivity, in and out degree of each dyad member, and
network density—and obtain estimators of the effect of these parameters
on the conditional probability of {x

ij

⫽ 1}. Our models include network

parameters that are similar to the ones of a typical P* model—transitive
triads, the degree of each dyad member, and network density—but we
measured these parameters on the network at (t

⫺ 1), while a strict pseu-

dolikelihood estimation requires parameters measured on the same net-
work that contains the predicted tie. Since the inclusion of the (t

⫺ 1)

parameters cannot be considered an adequate safeguard against the po-
tential effects of nonindependent observations, we used the above-
mentioned MRQAP-like procedure to test the robustness of the results
and limit concerns of interdependence. The percentage of frequency with
which the results in the random-sample simulations exceeded the original
estimates was far less than 5% in all instances. Thus, we can say with
some confidence that for these data reasonable coefficients were obtained.

The problem of cross-sectional dyadic interdependence can also be un-

derstood as one of model misspecification (Lincoln 1984). If a statistical
model incorporated all essential nodal (organization-level) characteristics
that influence alliance formation, no unobserved effects resulting from
common nodes would remain. To capture any organization-level effects
across dyads sharing the same organization, we controlled for each organi-
zation’s cumulative history of alliances. Organization history is an impor-
tant factor that captures any residual organizational propensities to en-
gage in alliances (Heckman and Borjas 1980; Black, Moffitt, and Warner
1990). As noted earlier, we also ran separate estimations in which we in-
cluded a host of financial attributes of each of the organizations in the
dyad, including its size, performance, liquidity, and solvency. In addition
to these controls, the models used here account for unobserved heterogene-
ity and adjust for such systematic biases resulting from missing variables.
We expected the unobserved heterogeneity term (

ρ

) to capture any resid-

ual dyad-level effects not included in the model.

1483

American Journal of Sociology

Unobserved Heterogeneity

An issue that arises when analyzing data on a time series of cross-sections,
or panel data, is the possibility of unobserved time-invariant effects
known as “unobserved heterogeneity.” This is of particular concern for
this study with respect to the claim that the prior history of alliances be-
tween two organizations affects the future likelihood of their entering an
alliance. There are two distinct explanations for this empirical regularity,
if it occurs (Heckman 1981a, 1981b). One explanation is that a genuine
behavioral effect exists, whereby, because of the prior alliances it has expe-
rienced, a dyad’s preferences are altered in the future. In econometric
terms, such a behavioral effect is called “state dependence”—the likeli-
hood of an event is a function of the state of the unit.

If state dependence alone encapsulated the empirical reality, there

would be no problem; however, there is another possibility that, if not
accounted for, could lead to spurious results: dyads may differ in their
propensity to enter alliances because of unobserved factors. In this in-
stance, such unobservable effects could result from permanent differences
between dyads in their preferences for alliances, such as geographical
proximity, not captured by the independent variables. If this noise were
systematic for the same unit over time, it could lead to a serial correlation
among the error terms for those observations, which would yield consis-
tent but inefficient coefficients, rendering any statistical testing inaccurate.
Furthermore, prior alliance experience may appear to be a determinant
of future alliance formation solely because it is a proxy for temporally
persistent unobservable factors that determine alliance formation and
nonformation. Improper treatment can lead to spurious effects appearing
with attempts to assess the influence of past experience on current deci-
sions; this phenomenon is also termed “spurious state dependence” (Black
et al. 1990; Heckman 1981a, 1981b; Hsiao 1986).

In a statistical sense, the problem of unobserved heterogeneity relates

to model specification (Peterson and Koput 1991). If a model is completely
specified, no such problem occurs, but most statistical models suffer from
some degree of omitted variable bias. Another way to confront this prob-
lem is to refine the risk set studied. In the current design, we include all
possible dyads within each industry for each year as the set of dyads at
risk of entering an alliance. It is quite likely that some of these dyads are
in fact not at risk of entering an alliance in some or even all observation
periods, while other dyads have a higher propensity to ally. This suggests
the possibility of misspecification of the risk set unless adequate allow-
ances are made for such unobserved differences in propensity. One way
to deal with such a bias is to clean up the risk set by eliminating records

1484

Networks

unlikely to experience the event, a process analogous to removing men
from pregnancy studies. The difference in propensity is frequently a result
of unobservable factors, however, making it impossible a priori to weed
out records from the sample on reasonable grounds without biasing the
sample.

Two approaches frequently used to address problems of unobserved

heterogeneity are fixed- and random-effects models. Fixed-effects models
treat the unobserved individual effect as a constant over time and com-
pute it for each unit (dyad). The method entails estimating a constant
term for each distinct unit and including dummy variables for each and is
similar to least squares with dummy variables (LSDV) regression models
(Hannan and Young 1977; Mizruchi 1989). Random-effects models treat
the heterogeneity that varies across units as randomly drawn from some
underlying probability distribution. Both types of models have shortcom-
ings. Both assume that the unobserved effects are time invariant. Fixed-
effects models are applicable only to repeatable events (Yamaguchi 1991),
do not allow the inclusion of time-independent covariates ( Judge et al.
1985; Reader 1993), and involve estimating a large number of parameters,
which grows with sample size (Chamberlain 1985). This approach can be
problematic when there are many groups but only a few observations in
each group (Chamberlain 1985). Random-effects models are more tracta-
ble but also assume that the unobserved effect is not correlated with any
of the exogenous variables in the system (Chintagunta, Jain, and Vilcassim
1991; Hausman and McFaden 1984).

To address concerns of heterogeneity, we employed a random-effects

panel probit model, developed by Butler and Moffitt (1982), for the statis-
tical analysis.

16

Our decision to employ a random-effects model was based

on the following. First, estimates computed using fixed-effects models can
be biased for panels over short periods (Chintagunta et al. 1991; Heckman
1981a, 1981b; Hsiao 1986). This is not a problem with random-effects
models. As all the dyads in our sample were present for only nine years,
random effects was clearly the favored approach. Second, fixed-effects
models cannot include time-independent covariates, a limitation that
would have meant excluding several variables, and an analysis without
some of these variables would have been severely limited. The computa-

16

In random-effects models, numerous alternatives are possible, depending on the

choice of form for the distribution of unobservables. Although Butler and Moffitt
(1982) specified a normal distribution, other functional forms are also possible. Recent
efforts have moved away from functional specification of heterogeneity toward semi-
parametric random-effects approaches that estimate the probability distribution di-
rectly from the data (cf. Chintagunta et al. 1991).

1485

American Journal of Sociology

tion of random-effects models is relatively straightforward for continuous
dependent variables but more problematic for qualitative choice variables
and was implemented here using LIMDEP 6.0.

We also tried to address concerns of heterogeneity by conducting the

analysis using three increasingly restrictive definitions of the risk set. The
first set included all dyads in the sample, the second set included only
dyads in which at least one member had prior alliance experience, and
the third set included dyads in which both members had entered into at
least one alliance. The results obtained with different sets were conver-
gent, and we report those based on the complete set.

Comparative Analyses

The primary theoretical contention underlying our use of network mea-
sures is that the ties formed in an industry are not random but are driven
by the structure of relationships formed in prior years. The models that
include network variables were expected to be powerful predictors of alli-
ance formation to the extent that (a) alliance formation among organiza-
tions arises from the flow of information underlying the networks of preex-
isting relationships and (b) the specific structural models used to reflect
these information flows cluster organizations that are densely connected
by such informational links (Friedkin 1984).

To verify our claims of systematic interorganizational alliances, we

compared the results for this study’s sample against results obtained with
a sample in which the formation of alliances was assigned randomly. The
implicit null hypothesis here is that an observed pattern in the data is due
purely to chance. Such a comparative analysis serves as a valuable base-
line (cf. Zajac 1988). Finding no differences in the predictive power of the
independent variables for the actual and random dependent variables, or
greater predictive power for the random dependent variable, would sug-
gest that the postulated independent effects could have predicted the ran-
dom occurrence of alliances just as well or better. As a result, our claims
for systematic patterning of alliances would be moot.

We tested the predictive ability of each model specified in table 3

against random assignments on the dependent variable on the basis of its
original distribution. The results indicated that none of the hypothesized
effects are better predictors of randomly assigned alliances than those in
table 3. Not a single independent variable is significant in all the models.
This finding allows us to reject the implicit null hypothesis and suggests
that the postulated independent effects are not at all good predictors of
the random occurrence of alliances. The exogenous interdependence and
endogenous embeddedness effects explain the systematic pattern of alli-
ances.

1486

Networks

REFERENCES

Aiken, Michael, and Jerald Hage. 1968. “Organizational interdependence and Intraor-

ganizational Structure.” American Sociological Review 33:912–30.

Alba, Richard D., and Charles Kadushin. 1976. “The Intersection of Social Circles:

A New Measure of Social Proximity in Networks.” Sociological Methods and Re-
search 5:77–102.

Alba, Richard D., and Gwen Moore. 1983. “Elite Social Circles.” Pp. 245–61 in Applied

Network Analysis, edited by Ronald S. Burt, Michael Minor, and Associates. Bev-
erly Hills, Calif.: Sage.

Aldrich, Howard. 1979. Organizations and Environments. Englewood Cliffs, N.J.:

Prentice-Hall.

Arthur, Brian W. 1989. “Competing Technologies and Lock-In by Historical Events.”

Economic Journal 99 (394): 116–31.

Baker, Wayne E. 1990. “Market Networks and Corporate Behavior.” American Jour-

nal of Sociology 96:589–625.

Barley, S. R., J. Freeman, and R. C. Hybels. 1992. “Strategic Alliances in Commercial

Biotechnology.” Pp. 311–47 in Networks and Organizations: Structure, Form and
Action, edited by N. Nohria and R. Eccles. Boston: Harvard Business School Press.

Barney, J. B., and R. E. Hoskisson. 1990. “Strategic Groups: Untested Assertions and

Research Proposals.” Managerial and Decision Economics 11:187–98.

Berg, S., and P. Friedman. 1980. “Causes and Effects of Joint Venture Activity.” Anti-

trust Bulletin 25:143–68.

Black, M., R. Moffitt, and J. T. Warner. 1990. “The Dynamics of Job Separation: The

Case of Federal Employees.” Journal of Applied Econometrics 5:245–62.

Blau, P. 1977. “A Macrosociological Theory of Social Structure.” American Journal

of Sociology 83:26–54.

Bonacich, Phillip. 1987. “Power and Centrality: A Family of Measures.” American

Journal of Sociology 92:1170–82.

Borgatti, S. 1988. “A Comment on Doreian’s Regular Equivalence in Symmetric

Structures.” Social Networks 10:265–71.

Borgatti, S. P., and M. G. Everett. 1994. “Notions of Position in Social Network Analy-

sis.” Sociological Methodology, vol. 22. San Francisco: Jossey-Bass.

Borgatti, S. P., M. G. Everett, and L. C. Freeman. 1992. UCINET IV. Analytic Tech-

nologies, Columbia, S.C.

Bradach, J. L., and R. G. Eccles. 1989. “Price, Authority, and Trust: From Ideal Types

to Plural Forms.” Annual Review of Sociology 15:97–118.

Burt, R. S. 1976. “Positions in Networks.” Social Forces 55:93–112.
———. 1982. Toward a Structural Theory of Action. New York: Academic Press.
———. 1983. Corporate Profits and Cooptation: Networks of Market Constraints and

Directorate Ties in the American Economy. New York: Academic Press.

———. 1987. “Social Contagion and Innovation: Cohesion versus Structural Equiva-

lence.” American Journal of Sociology 92:1287–1335.

———. 1988. “The Stability of American Markets.” American Journal of Sociology

93:356–95.

———. 1990. “Detecting Role Equivalence.” Social Networks 12:83–97.
———. 1991. Structure Ver. 4.2. Center for the Social Sciences, Columbia University,

New York.

———. 1992. Structural Holes: The Social Structure of Competition. Cambridge,

Mass.: Harvard University Press.

Burt, R. S., and Marc Knez. 1995. “Kinds of Third-Party Effects on Trust.” Rational-

ity and Society 7:225–92.

Butler, J. S., and Robert Moffitt. 1982. “A Computationally Efficient Quadrature Pro-

cedure for the One-Factor Multinomial Probit Model.” Econometrica 50:761–64.

1487

American Journal of Sociology

Carley, K. 1990. “Group Stability: A Socio-Cognitive Approach.” Advances in Group

Processes 7:1–44.

———. 1991. “A Theory of Group Stability.” American Sociological Review 56:331–

54.

Chamberlain, Gary. 1985. “Heterogeneity, Omitted Variable Bias, and Duration De-

pendence.” Pp. 3–38 in Longitudinal Analysis of Labor Market Data, edited by J.
Heckman and B. Singer. New York: Cambridge University Press.

Chintagunta, Pradeep K., Dipak C. Jain, and Naufel J. Vilcassim. 1991. “Investigating

Heterogeneity in Brand Preference in Logit Models for Panel Data.” Journal of
Marketing Research 27:417–28.

Coase, R. H. (1937) 1952. “The Nature of the Firm.” Pp. 331–51 in A.E.A. Readings

in Price Theory, edited by G. J. Stigler and K. E. Boulding. Homewood, Ill.: Irwin.

Coleman, James C. 1990. Foundations of Social Theory. Cambridge, Mass.: Harvard

University Press.

Contractor, F., and P. Lorange. 1988. Cooperative Strategies in International Busi-

ness. Lexington, Mass.: Lexington Books.

Davis, Gerald F., Kristina Diekmann, and Catherine H. Tinsley. 1994. “The Decline

and Fall of the Conglomerate Firm in the 1980s: The Deinstitutionalization of an
Organizational Form.” American Sociological Review 59:547–70.

Davis, Gerald F., and Henrich R. Greve. 1997. “Corporate Elite Networks and Gover-

nance Changes in the 1980s.” American Journal of Sociology 103 (1): 1–37.

Dooley, Peter C. 1969. “The Interlocking Directorate.” American Economic Review

59:314–23.

Dore, R. 1983. “Goodwill and the Spirit of Market Capitalism.” British Journal of

Sociology 34:459–82.

Doreian, P. 1987. “Measuring Equivalence in Symmetric Structures.” Social Networks

9:89–107.

———. 1988. “Borgatti Toppings on Doreian Splits: Reflections on Regular Equiva-

lence.” Social Networks 10:273–85.

Doz, Y. L. 1996. “The Evolution of Cooperation in Strategic Alliances: Initial Condi-

tions or Learning Process?” Strategic Management Journal 17:55–83.

Doz, Y., Gary Hamel, and C. K. Prahalad. 1989. “Collaborate with Your Competitors

and Win.” Harvard Business Review 67:133–39.

Duncan, L. 1982. “Impacts of New Entry and Horizontal Joint Ventures on Industrial

Rates of Return.” Review of Economics and Statistics 64:120–25.

Durkheim, Emile. (1893) 1933. The Division of Labor in Society. New York: Free

Press.

———. (1897) 1951. Suicide. New York: Free Press.
Eccles, Robert G. 1981. “The Quasifirm in the Construction Industry.” Journal of

Economic Behavior and Organization 2:335–57.

Faust, K. 1988. “Comparison of Methods for Positional Analysis: Structural and Gen-

eral Equivalences.” Social Networks 10:313–41.

Fernandez, Roberto M. 1991. “Structural Bases of Leadership in Intraorganizational

Networks.” Social Psychology Quarterly 54:36–53.

Freeman, L. C. 1979. “Centrality in Social Networks: Conceptual Clarification.” So-

cial Networks 1:215–39.

Friedkin, Noah E. 1984. “Structural Cohesion and Equivalence Explanations of Social

Homogeneity.” Sociological Methods and Research 12:235–61.

Galaskiewicz, J. 1982. “Modes of Resource Allocation: Corporate Contributions to

Nonprofit Organizations.” Pp. 235–53 in Social Structure and Network Analysis,
edited by Peter V. Marsden and Nan Lin. Beverly Hills, Calif.: Sage.

———. 1985. “Inter-Organizational Relations.” Pp. 281–304 in Annual Review of So-

ciology, vol. 11. Edited by Ralph H. Turner and James F. Short, Jr. Palo Alto,
Calif.: Annual Reviews.

1488

Networks

Galaskiewicz, J., and R. S. Burt. 1991. “Interorganization Contagion in Corporate

Philanthropy.” Administrative Science Quarterly 36:88–105.

Gargiulo, M. 1993. “Two-Step Leverage: Managing Constraint in Organizational Poli-

tics.” Administrative Science Quarterly 38:1–19.

———. 1998. “Structure and Action: A Network Approach to the Micro-Macro Link.”

Working paper. INSEAD, Department of Organizational Behavior.

Gargiulo, M., and M. Benassi. 1998. “The Dark Side of Social Capital.” In Social

Capital and Liability, edited by Shaul Gabbay and Roger Leenders. Norwell, Mass.:
Kluver. In press.

Ghemawat, P., M. Porter, and R. Rawlinson. 1986. “Patterns of International Coali-

tion Activity.” Pp. 345–66 in Competition in Global Industries, edited by M. Porter.
Boston: Harvard Business School Press.

Granovetter, Mark. 1985. “Economic Action and Social Structure: A Theory of Em-

beddedness.” American Journal of Sociology 91:481–510.

———. 1992. “Problems of Explanation in Economic Sociology.” Pp. 25–56 in Net-

works and Organizations: Structure, Form and Action, edited by N. Nohria and R.
Eccles. Boston: Harvard Business School Press.

Gulati, R. 1995a. “Familiarity Breeds Trust? The Implications of Repeated Ties on

Contractual Choice in Alliances.” Academy of Management Journal 38:85–112.

———. 1995b. “Social Structure and Alliance Formation Pattern: A Longitudinal

Analysis.” Administrative Science Quarterly 40:619–52.

———. 1998. “Alliances and Networks.” Strategic Management Journal 19:293–

317.

———. 1999. “Network Location and Learning: The Influence of Network Resources

and Firm Capabilities on Alliance Formation.” Strategic Management Journal, in
press.

Gulati, R., and P. Lawrence. 1999. “The Diversity of Embedded Ties.” Working paper.

J. L. Kellogg, Graduate School of Management.

Gulati, R., and H. Singh. 1998. “The Architecture of Cooperation: Managing Coordi-

nation Uncertainty and Interdependence in Strategic Alliances.” Administrative
Science Quarterly, 43:781–814.

Gulati, R., and J. Westphal. 1999. “The Dark Side of Embeddedness: An Examination

of the Influence of Direct and Indirect Board Interlocks and CEO/Board Relation-
ships on Interfirm Alliances.” Administrative Science Quarterly, in press.

Hallinan, M. T., and A. B. Sorensen. 1985. “Ability Grouping and Student Friend-

ships.” American Educational Research Journal 22:485–99.

Hamilton, Gary, and Nicole Biggart. 1988. “Market, Culture, and Authority.” Ameri-

can Journal of Sociology 94 (suppl.): S52–S94.

Han, Shin-Kap. 1994. “Mimetic Isomorphism and Its Effects on the Audit Service

Market.” Social Forces 73:637–63.

Hannan, Michael T., and John Freeman. 1989. Organizational Ecology. Cambridge,

Mass.: Harvard University Press.

Hannan, Michael, and Alice Young. 1977. “Estimation in Panel Models: Results on

Pooling Cross-Sections and Time Series.” Pp. 52–83 in Sociological Methodology,
edited by Samuel Leinhardt. San Francisco: Jossey-Bass.

Harrigan, Kathryn R. 1986. Managing for Joint Ventures Success. Lexington, Mass.:

Lexington Books.

Hausman, J. A., and D. L. McFadden. 1984. “Specification Tests for the Multinomial

Logit Model.” Econometrica 52:1219–40.

Haveman, Heather A., and Hayagreeva Rao. 1997. “Structuring a Theory of Moral

Sentiments: Institutional and Organizational Coevolution in the Early Thrift Indus-
try.” American Journal of Sociology 102 (6): 1606–51.

Heckman, J. 1981a. “Heterogeneity and State Dependence” Pp. 91–139 in Studies in

Labor Markets, edited by S. Rosen. Chicago: University of Chicago Press.

1489

American Journal of Sociology

———. 1981b. “Statistical Models for Discrete Panel Data.” In The Econometrics of

Panel Data, edited by D. McFadden and C. Manski. Cambridge, Mass.: MIT Press.

Heckman, J., and G. J. Borjas. 1980. “Does Unemployment Cause Future Unemploy-

ment: Definitions, Questions, and Answers from a Continuous Time Model of Het-
erogeneity and State Dependence.” Econometrica 47:247–83.

Hergert, M., and D. Morris. 1988. “Trends in International Collaborative Agree-

ments.” Pp. 99–110 in Cooperative Strategies in International Business, edited by
F. K. Contractor and P. Lorange. Lexington, Mass.: Lexington Books.

Holland, P. W., and S. Leinhardt. 1970. “A Method for Detecting Structure in Socio-

metric Data.” American Journal of Sociology 76:492–513.

Hsiao, Cheng. 1986. Analysis of Panel Data. New York: Cambridge University Press.
Hummel, H. J., and W. Sodeur. 1987. “Struckturbeschreibung von Positionen in sozia-

len Beziehungsnetzen.” Methoden der Netwerkanalyze. Munich: F. U. Pappi.

Iacobucci, D., and S. Wasserman. 1988. “A General Framework for the Statistical

Analysis of Sequential Dyadic Interaction Data.” Psychological Bulletin 103 (3):
379–90.

Jaccard, James, Robert Turrisi, and Choi K. Wan. 1990. Interaction Effects in Multi-

ple Regression. London: Sage.

Judge, George G., W. E. Griffiths, R. C. Hill, and Tsoung-Chao Lee. 1985. The Theory

and Practice of Econometrics. New York: Wiley.

Kogut, Bruce. 1988. “Joint Ventures: Theoretical and Empirical Perspectives.” Strate-

gic Management Journal 9:319–32.

Kogut, Bruce, Weijan Shan, and Gordon Walker. 1992. “Competitive Cooperation in

Biotechnology: Learning through Networks?” Pp. 348–65 in Networks and Organi-
zations: Structure, Form, and Action, edited by N. Nohria and R. Eccles. Boston:
Harvard Business School Press.

Krackhardt, David. 1987. “QAP Partialling as a Test for Spuriousness.” Social Net-

works 9:171–86.

———. 1988. “Predicting with Networks: Nonparametric Multiple Regression Analy-

sis of Dyadic Data.” Social Networks 10:359–81.

———. 1990. “Assessing the Political Landscape: Structure, Cognition and Power in

Organizations.” Administrative Science Quarterly 35:342–69.

Leenders, R. T. A. J. 1995. “Models for Network Dynamics: A Markovian Frame-

work.” Journal of Mathematical Sociology 20:1–21.

———. 1996. “Evolution of Friendship and Best Friendship Choices.” Journal of

Mathematical Sociology 21:133–48.

Leifer, Eric, and Harrison C. White. 1988. “A Structural Approach to Markets.” Pp.

85–108 in Intercorporate Relations: The Structural Analysis of Business, edited by
Mark S. Mizruchi and Michael Schwartz. New York: Cambridge University Press.

Lincoln, James R. 1984. “Analyzing Relations in Dyads.” Sociological Methods and

Research 13:45–76.

Lincoln, J. R., M. L. Gerlach, C. L. Ahmadjian. 1996. “Keiretsu Networks and Corpo-

rate Performance in Japan.” American Sociological Review 61:67–88.

Linton, Ralph. 1936. The Study of Man. New York: D. Appleton-Century.
Little, Roderick J., and Donald B. Rubin. 1987. Statistical Analysis with Missing

Data. New York: Wiley.

Lorrain, Francois, and Harrison C. White. 1971. “Structural Equivalence of Individu-

als in Social Networks.” Journal of Mathematical Sociology 1:49–80.

Macauley, S. 1963. “Non-Contractual Relations in Business: A Preliminary Study.”

American Sociological Review 28:55–67.

MacIntyre, Alasdair. 1981. After Virtue. Notre Dame, Ind.: University of Notre Dame

Press.

Manley, Bryan F. 1992. The Design and Analysis of Research Studies. New York:

Cambridge University Press.

1490

Networks

Mariti, P. and R. H. Smiley. 1983. “Co-operative Agreements and the Organization

of Industry.” Journal of Industrial Economics 31:437–51.

Marsden, Peter V., and Noah E. Friedkin. 1993. “Network Studies of Social Influ-

ence.” Sociological Methods and Research 22:127–51.

Merton, Robert K. 1957. “The Role-Set: Problems in Sociological Theory.” British

Journal of Sociology 8:106–20.

Mizruchi, Mark S. 1989. “Similarity of Political Behavior among Large American

Corporations.” American Journal of Sociology 95:401–24.

———. 1992. The Structure of Corporate Political Action: Interfirm Relationships

and Their Consequences. Cambridge, Mass.: Harvard University Press.

———. 1993. “Cohesion, Equivalence, and Similarity of Behavior: A Theoretical and

Empirical Assessment.” Social Networks 15:275–308.

Mizruchi, Mark S., and Linda B. Stearns. 1988. “A Longitudinal Study of the Forma-

tion of Interlocking Directorates.” Administrative Science Quarterly 33:194–210.

Nadel, Siegfried F. 1957. The Theory of the Social Structure. London: Cohen and

West.

Nohria, N., and C. Garcia-Pont. 1991. “Global Strategic Linkages and Industry Struc-

ture.” Strategic Management Journal 12:105–24.

Oliver, Christine. 1990. “Determinants of Inter-Organizational Relationships: Integra-

tion and Future Directions.” Academy of Management Review 15:241–65.

Palmer, D., R. Friedland, and J. V. Singh. 1986. “Stability in a Corporate Interlock

Network.” American Sociological Review 51:781–96.

Peterson, Trond, and Ken Koput. 1991. “Density Dependence in Organizational Mor-

tality: Legitimacy or Unobserved Heterogeneity?” American Sociological Review
56:399–409.

Pfeffer, J., and P. Nowak. 1976a. “Joint Venture and Interorganizational Interdepen-

dence.” Administrative Science Quarterly 21:398–418.

———. 1976b. “Patterns of Joint Venture Activity: Implications for Anti-Trust Re-

search.” Antitrust Bulletin 21:315–39.

Pfeffer, J., and Gerald Salancik. 1978. The External Control of Organizations: A Re-

source Dependence Perspective. New York: Harper and Row.

Podolny, Joel M. 1993. “A Status-Based Model of Market Competition.” American

Journal of Sociology 98:829–72.

———. 1994. “Market Uncertainty and the Social Character of Economic Exchange.”

Administrative Science Quarterly 39:458–83.

Podolny, Joel M., and Toby E. Stuart. 1995. “A Role-Based Ecology of Technological

Change.” American Journal of Sociology 100 (5): 1224–60.

Popielarz, Pamela A., and J. Miller McPherson. 1995. “On the Edge or in Between:

Niche Position, Niche Overlap, and the Duration of Voluntary Association Mem-
berships.” American Journal of Sociology 101 (3): 698–720.

Porter, M. E. 1979. “The Structure within Industries and Companies’ Performance.”

Review of Economics and Statistics 61:214–27.

———. 1990. The Competitive Advantage of Nations. New York: Free Press.
Portes, A., and J. Sensenbrenner. 1993. “Embeddedness and Immigration: Notes on

the Social Determinants of Economic Action.” American Journal of Sociology 98:
1320–50.

Powell, W. W. 1990. “Neither Market nor Hierarchy: Network Forms of Organiza-

tion.” Research in Organizational Behavior 12:295–336.

Powell, Walter W., Kenneth Koput, and Laurel Smith-Doerr. 1996. “Inter-Organiza-

tional Collaboration and the Locus of Innovation: Networks of Learning in Biotech-
nology.” Administrative Science Quarterly 41:116–45.

Powell, W. W., and Laurel Smith-Doerr. 1994. “Networks and Economic Life.” Pp.

368–402 in The Handbook of Economic Sociology, edited by Neil Smelser and Rich-
ard Swedberg. Princeton, N.J.: Princeton University Press.

1491

American Journal of Sociology

Raub, Werner, and J. Weesie. 1990. “Reputation and Efficiency in Social Interactions:

An Example of Network Effects.” American Journal of Sociology 96 (3): 626–54.

Reader, S. 1993. “Unobserved Heterogeneity in Dynamic Discrete Choice Models.”

Environment and Planning 25:495–519.

Reger, R. K., and A. S. Huff. 1993. “Strategic Groups: A Cognitive Perspective.” Stra-

tegic Management Journal 14:103–24.

Ring, P. S., and Andrew H. Van de Ven. 1992. “Structuring Cooperative Relationships

between Organizations.” Strategic Management Journal 13:483–98.

Romo, Frank P., and Michael Schwartz. 1995. “The Structural Embededdness of

Business Decisions: The Migration of Manufacturing Plants in New York State,
1960 to 1985.” American Sociological Review 60:874–907.

Scott, W. Richard, 1995. Institutions and Organizations. London: Sage.
Shan, W., and W. Hamilton. 1991. “Country-Specific Advantage and International

Cooperation.” Strategic Management Journal 12:419–32.

Snijders, T. A. 1996. “Stochastic Actor-Oriented Model for Network Change.” Journal

of Mathematical Sociology 21:149–72.

Stern, Robert N. 1979. “The Development of an Interorganizational Control Network:

The Case of Intercollegiate Athletics.” Administrative Science Quarterly 24 (2):
242–66.

Stinchcombe, Arthur L. 1990. Information and Organization. Berkeley and Los

Angeles: University of California Press.

Strauss, D., and M. Ikeda. 1990. “Pseudolikelihood Estimation for Social Networks.”

Journal of the American Statistical Association 85:204–12.

Swedberg, Richard. 1994. “Markets as Social Structures.” Pp. 255–82 in The Hand-

book of Economic Sociology, edited by Neil Smelser and Richard Swedberg.
Princeton, N.J.: Princeton University Press.

Thomas, Howard, and N. Venkatraman. 1988. “Research in Strategic Groups: Prog-

ress and Prognosis.” Journal of Management Studies 25:537–55.

Useem, M. 1984. The Inner Circle. New York: Oxford University Press.
Van de Ven, Andrew H. 1976. “On the Nature, Formation and Maintenance of Rela-

tions among Organizations.” Academy of Management Review 1:24–36.

Van Rossem, R. 1996. “The World System Paradigm as a General Theory of Develop-

ment: A Cross-National Test.” American Sociological Review 61:508–27.

Venkatraman, N., L. Loh, and J. Koh. 1994. “The Adoption of Corporate Governance

Mechanisms: A Text of Competing Diffusion Models.” Management Science 40 (4):
496–507.

Walker, G., B. Kogut, W. Shan. 1997. “Social Capital, Structural Holes, and the For-

mation of an Industry Network.” Organization Science 8:109–25.

Wasserman, Stanley, and Katherine Faust. 1994. Social Network Analysis: Methods

and Applications. Cambridge: Cambridge University Press.

Wasserman, S., and P. Pattison 1996. “Logit Models and Logistic Regressions for Uni-

variate and Bivariate Social Networks: I. An Introduction to Markov Graphs and
p*.” Psychometrika 61:401–26.

Wellman, Barry. 1988. “Structural Analysis: From Method and Metaphor to Theory

and Substance.” Pp. 19–61 in Social Structures: A Network Approach, edited by
B. Wellman and D. Berkowitz. Cambridge: Cambridge University Press.

Westphal, James, Ranjay Gulati, and Steve Shortell. 1997. “An Institutional and Net-

work Perspective on the Content and Consequences of TQM Adoption.” Adminis-
trative Science Quarterly 42:366–94.

White, Harrison C. 1981. “Where Do Markets Come From?” American Journal of

Sociology 87:517–47.

White, H. C., S. A. Boorman, and R. Breiger. 1976. “Social Structure from Multiple

Networks: I. Blockmodels of Roles and Positions.” American Journal of Sociology
81:730–80.

1492

Networks

Williamson, Oliver. 1985. The Economic Institutions of Capitalism. New York: Free

Press.

Winship, Christopher, and Michael Mandel. 1983. “Roles and Positions: A Critique

and Extension of the Blockmodeling Approach.” Pp. 314–44 in Sociological Meth-
odology, edited by Samuel Leinhardt.

Wippler, Reinhard, and Siegwart Lindenberg. 1987. “Collective Phenomena and Ra-

tional Choice.” Pp. 135–52 in The Micro-Macro Link, edited by Jeffrey C. Alexander
et al. Berkeley and Los Angeles: University of California Press.

Yamaguchi, Kazuo. 1991. Event History Analysis. Newbury Park, Calif.: Sage.
Zaheer, A., B. McEvily, and V. Perrone. 1998. “Does Trust Matter? Exploring the

Effects of Interorganizational and Interpersonal Trust on Performance.” Organiza-
tion Science 9:1–20.

Zajac, Edward J. 1988. “Interlocking Directorates as an Inter-Organizational Strat-

egy: A Test of Critical Assumptions.” Academy of Management Journal 31:428–38.

Zeggelink, E. 1994. “Dynamics of Structure: An Individual Oriented Approach.” So-

cial Networks 16:295–333.

1493